[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17335172#comment-17335172 ] Tsz-wo Sze commented on HDFS-7285: -- Below is https://img-ask.csdnimg.cn/upload/1619363340018.png : !1619363340018.png! > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang >Priority: Major > Fix For: 3.0.0-alpha1 > > Attachments: 1619363340018.png, Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-Consolidated-20150911.patch, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17335169#comment-17335169 ] Stone commented on HDFS-7285: - [~zhz] https://img-ask.csdnimg.cn/upload/1619363340018.png You can see this picture by coping the URL address to the browser.Don't open this url link directly > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang >Priority: Major > Fix For: 3.0.0-alpha1 > > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-Consolidated-20150911.patch, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17332809#comment-17332809 ] Zhe Zhang commented on HDFS-7285: - [~iostream2...@163.com] Unfortunately I don't have permission to the [link|[https://img-ask.csdnimg.cn/upload/1619363340018.png]] you just pasted. It shows "403 Forbidden". Could you find a way to upload the image to an accessible location, or describe the issue? Thanks > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang >Priority: Major > Fix For: 3.0.0-alpha1 > > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-Consolidated-20150911.patch, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17331776#comment-17331776 ] Stone commented on HDFS-7285: - [~zhz] I have no permission to upload attachment.Please visit csdn web the url of image->[https://img-ask.csdnimg.cn/upload/1619363340018.png] > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang >Priority: Major > Fix For: 3.0.0-alpha1 > > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-Consolidated-20150911.patch, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17327055#comment-17327055 ] Zhe Zhang commented on HDFS-7285: - [~iostream2...@163.com]: the image cannot be viewed. Can you upload again? Thanks > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang >Priority: Major > Fix For: 3.0.0-alpha1 > > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-Consolidated-20150911.patch, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17325648#comment-17325648 ] Stone commented on HDFS-7285: - !image-2021-04-20-17-26-45-834.png! I can not understand it.For example,if 2-missing,the value of column named 3-replication is 1LR+1RR.Why not 1LN+1RR?What it means? > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang >Priority: Major > Fix For: 3.0.0-alpha1 > > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-Consolidated-20150911.patch, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15897911#comment-15897911 ] Andrew Wang commented on HDFS-7285: --- Hi Marcell, I've filed HDFS-11503 so we can continue discussion off of this umbrella JIRA with many watchers. I'll respond to your above comment over there. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Fix For: 3.0.0-alpha1 > > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > fsimage-analysis-20150105.pdf, HDFS-7285-Consolidated-20150911.patch, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, HDFS-bistriped.patch, > HDFS-EC-merge-consolidated-01.patch, HDFS-EC-Merge-PoC-20150624.patch, > HDFSErasureCodingDesign-20141028.pdf, HDFSErasureCodingDesign-20141217.pdf, > HDFSErasureCodingDesign-20150204.pdf, HDFSErasureCodingDesign-20150206.pdf, > HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.15#6346) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15893989#comment-15893989 ] Marcell Feher commented on HDFS-7285: - Hi Andrew! Alright, we'll use the "trunk". Yes, the X axis is the cell/chunk size. These charts reflect preliminary speeds that we quickly made just to see an approximate comparison against ISA-L. We are currently working on a more reliable benchmark, I'll post the results when they are in. Theoretically our encode speed is mostly dominated by the number of data units and reconstruct speed by the number of lost parity units that needs to be repaired. Today we are not compatible with ISA and the built-in plugins (since our plugin uses the Vandermonde coefficient matrix), but we identified this problem as well and working actively on the Cauchy matrix implementation. Our goal is to be compatible and we will reach in the next few weeks. Our RS plugin, similarly to ISA-L, is implemented in C++ and offered as a native binary library. The only thing needs to be shipped within the Hadoop framework is the glue code. Our understanding is that this is the intended way of shipping new native EC plugins. Some glue code is always necessary in the framework (like with Liberasurecode). Isn't this the case? > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Fix For: 3.0.0-alpha1 > > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > fsimage-analysis-20150105.pdf, HDFS-7285-Consolidated-20150911.patch, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, HDFS-bistriped.patch, > HDFS-EC-merge-consolidated-01.patch, HDFS-EC-Merge-PoC-20150624.patch, > HDFSErasureCodingDesign-20141028.pdf, HDFSErasureCodingDesign-20141217.pdf, > HDFSErasureCodingDesign-20150204.pdf, HDFSErasureCodingDesign-20150206.pdf, > HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.15#6346) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15892889#comment-15892889 ] Andrew Wang commented on HDFS-7285: --- Hi Marcell, I recommend basing the patch off of the "trunk" branch, we've been pretty actively working on EC in preparation for alpha3. I looked at the benchmark results, very interesting! I'm guessing the X axis is the EC cell size? We've chosen the RS parameters for our built-in EC policies somewhat arbitrarily, but maybe we should tune them for better performance. Are there any patterns you see wrt the RS parameters? It looks like the #1 factor is how many parity blocks there are, though there's some weirdness like 10+4 being faster than 9+4. There seem to be differences based on the cell size, but no pattern I can see. Another question, is the Chocolate Cloud RS implementation byte-for-byte compatible with the Hadoop ISA-L and Java RS coder implementations? If not, it'll be harder for users to experiment with CC RS since it'll require rewriting data. We'll also need to add a new set of EC policies, which increases the configuration and maintenance overhead. In this case, I'd like to see more clear wins and a stable native code release before we put this in a release. In the meanwhile, we can also set up a development branch so you can work on the Hadoop integration. We've also made some progress toward pluggable EC policies and coder framework, though it's not done yet. If this gets finished, it'd be a nice way of making your RS implementation accessible to users without requiring it to be shipped with Hadoop itself. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Fix For: 3.0.0-alpha1 > > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > fsimage-analysis-20150105.pdf, HDFS-7285-Consolidated-20150911.patch, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, HDFS-bistriped.patch, > HDFS-EC-merge-consolidated-01.patch, HDFS-EC-Merge-PoC-20150624.patch, > HDFSErasureCodingDesign-20141028.pdf, HDFSErasureCodingDesign-20141217.pdf, > HDFSErasureCodingDesign-20150204.pdf, HDFSErasureCodingDesign-20150206.pdf, > HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.15#6346) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15892054#comment-15892054 ] Marcell Feher commented on HDFS-7285: - Thanks for the help guys, we'll proceed with creating a sub-task and submitting our patch. We should used 3.0.0-alpha2 release as the base, right? > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Fix For: 3.0.0-alpha1 > > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > fsimage-analysis-20150105.pdf, HDFS-7285-Consolidated-20150911.patch, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, HDFS-bistriped.patch, > HDFS-EC-merge-consolidated-01.patch, HDFS-EC-Merge-PoC-20150624.patch, > HDFSErasureCodingDesign-20141028.pdf, HDFSErasureCodingDesign-20141217.pdf, > HDFSErasureCodingDesign-20150204.pdf, HDFSErasureCodingDesign-20150206.pdf, > HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.15#6346) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15890866#comment-15890866 ] Wei-Chiu Chuang commented on HDFS-7285: --- The Hadoop EC framework is designed to be pluggable, however, the native EC codec loader assumes it uses ISA-L library (ErasureCodeNative.java), so that probably need to be changed. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Fix For: 3.0.0-alpha1 > > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > fsimage-analysis-20150105.pdf, HDFS-7285-Consolidated-20150911.patch, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, HDFS-bistriped.patch, > HDFS-EC-merge-consolidated-01.patch, HDFS-EC-Merge-PoC-20150624.patch, > HDFSErasureCodingDesign-20141028.pdf, HDFSErasureCodingDesign-20141217.pdf, > HDFSErasureCodingDesign-20150204.pdf, HDFSErasureCodingDesign-20150206.pdf, > HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.15#6346) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15890862#comment-15890862 ] Zhe Zhang commented on HDFS-7285: - Thanks [~sw0rdf1sh]. In addition to what Chris pointed out, you might also want to take a look at the discussions under HADOOP-11264; since your library is on the RS layer, sounds like the library can be plugged in with our coder framework. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Fix For: 3.0.0-alpha1 > > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > fsimage-analysis-20150105.pdf, HDFS-7285-Consolidated-20150911.patch, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, HDFS-bistriped.patch, > HDFS-EC-merge-consolidated-01.patch, HDFS-EC-Merge-PoC-20150624.patch, > HDFSErasureCodingDesign-20141028.pdf, HDFSErasureCodingDesign-20141217.pdf, > HDFSErasureCodingDesign-20150204.pdf, HDFSErasureCodingDesign-20150206.pdf, > HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.15#6346) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15890845#comment-15890845 ] Chris Douglas commented on HDFS-7285: - That sounds exciting, [~sw0rdf1sh]. There's a wiki [here|https://wiki.apache.org/hadoop/HowToContribute] that describes the patch process. If you need help please don't hesitate to reach out on [hdfs-dev|http://hadoop.apache.org/mailing_lists.html#HDFS]. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Fix For: 3.0.0-alpha1 > > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > fsimage-analysis-20150105.pdf, HDFS-7285-Consolidated-20150911.patch, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, HDFS-bistriped.patch, > HDFS-EC-merge-consolidated-01.patch, HDFS-EC-Merge-PoC-20150624.patch, > HDFSErasureCodingDesign-20141028.pdf, HDFSErasureCodingDesign-20141217.pdf, > HDFSErasureCodingDesign-20150204.pdf, HDFSErasureCodingDesign-20150206.pdf, > HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.15#6346) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15890254#comment-15890254 ] Marcell Feher commented on HDFS-7285: - Dear all! First of all let me introduce ourselves: we are Chocolate Cloud from Denmark, we use erasure coding to improve storage solutions. We already have Reed-Solomon and Random Linear Network Coding backends for Liberasurecode, and now we are at the final stage of developing our RS plugin to HDFS-EC. The performance of our plugin is similar to ISA-L's, in some configurations we are better, in others we are worse (our initial speed comparison charts can be found here: https://www.chocolate-cloud.cc/Plugins/HDFS-EC/hdfs.html). We would like our plugin to become officially supported in Hadoop 3.0. We can already provide a preliminary version of our (native) library and a patch with the necessary glue code for the next alpha release. I'd like to know your thoughts about whether it's possible and how it could be achieved. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Fix For: 3.0.0-alpha1 > > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > fsimage-analysis-20150105.pdf, HDFS-7285-Consolidated-20150911.patch, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, HDFS-bistriped.patch, > HDFS-EC-merge-consolidated-01.patch, HDFS-EC-Merge-PoC-20150624.patch, > HDFSErasureCodingDesign-20141028.pdf, HDFSErasureCodingDesign-20141217.pdf, > HDFSErasureCodingDesign-20150204.pdf, HDFSErasureCodingDesign-20150206.pdf, > HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.15#6346) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15861993#comment-15861993 ] Andrew Wang commented on HDFS-7285: --- Hi Doris, please see here for the current release plan: https://cwiki.apache.org/confluence/display/HADOOP/Hadoop+3.0.0+release If you're interested in EC, please do try out the recent 3.0.0-alpha2 release, we're actively working to improve the feature. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Fix For: 3.0.0-alpha1 > > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > fsimage-analysis-20150105.pdf, HDFS-7285-Consolidated-20150911.patch, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, HDFS-bistriped.patch, > HDFS-EC-merge-consolidated-01.patch, HDFS-EC-Merge-PoC-20150624.patch, > HDFSErasureCodingDesign-20141028.pdf, HDFSErasureCodingDesign-20141217.pdf, > HDFSErasureCodingDesign-20150204.pdf, HDFSErasureCodingDesign-20150206.pdf, > HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.15#6346) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15861265#comment-15861265 ] Doris Gu commented on HDFS-7285: Thanks a lot, Andrew. I see. Since such a situation, do you have any idea about the realese time of stable 3.0.0? > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Fix For: 3.0.0-alpha1 > > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > fsimage-analysis-20150105.pdf, HDFS-7285-Consolidated-20150911.patch, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, HDFS-bistriped.patch, > HDFS-EC-merge-consolidated-01.patch, HDFS-EC-Merge-PoC-20150624.patch, > HDFSErasureCodingDesign-20141028.pdf, HDFSErasureCodingDesign-20141217.pdf, > HDFSErasureCodingDesign-20150204.pdf, HDFSErasureCodingDesign-20150206.pdf, > HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.15#6346) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15859951#comment-15859951 ] Andrew Wang commented on HDFS-7285: --- Hi Doris, this patch was meant to apply to trunk. Getting this (and all the related EC patches) backported to 2.7 will be a lot of work, since 2.7 and trunk differ quite a bit. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Fix For: 3.0.0-alpha1 > > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > fsimage-analysis-20150105.pdf, HDFS-7285-Consolidated-20150911.patch, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, HDFS-bistriped.patch, > HDFS-EC-merge-consolidated-01.patch, HDFS-EC-Merge-PoC-20150624.patch, > HDFSErasureCodingDesign-20141028.pdf, HDFSErasureCodingDesign-20141217.pdf, > HDFSErasureCodingDesign-20150204.pdf, HDFSErasureCodingDesign-20150206.pdf, > HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.15#6346) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15859433#comment-15859433 ] Doris Gu commented on HDFS-7285: Hi there, I planed to apply this patch(Consolidated-20150810.patch),could anyone please tell me the branch which this patch can be applied to? And my goal is hadoopV2.7.3, is it possible to work? Looking forward to the answers. Best wishes! > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Fix For: 3.0.0-alpha1 > > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > fsimage-analysis-20150105.pdf, HDFS-7285-Consolidated-20150911.patch, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, HDFS-bistriped.patch, > HDFS-EC-merge-consolidated-01.patch, HDFS-EC-Merge-PoC-20150624.patch, > HDFSErasureCodingDesign-20141028.pdf, HDFSErasureCodingDesign-20141217.pdf, > HDFSErasureCodingDesign-20150204.pdf, HDFSErasureCodingDesign-20150206.pdf, > HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.15#6346) - To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14938646#comment-14938646 ] Hudson commented on HDFS-7285: -- FAILURE: Integrated in Hadoop-Mapreduce-trunk-Java8 #465 (See [https://builds.apache.org/job/Hadoop-Mapreduce-trunk-Java8/465/]) HDFS-8027. Erasure Coding: Update CHANGES-HDFS-7285.txt with branch (zhezhang: rev af8eaacdb40afba327b776e2b4f374be624289fc) * hadoop-hdfs-project/hadoop-hdfs/CHANGES-HDFS-EC-7285.txt HDFS-8027. Erasure Coding: Update CHANGES-HDFS-7285.txt with branch (zhezhang: rev 2c277802c1635837ec926862523be02e05e5649b) * hadoop-hdfs-project/hadoop-hdfs/CHANGES-HDFS-EC-7285.txt > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Fix For: 3.0.0 > > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-Consolidated-20150911.patch, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14938454#comment-14938454 ] Hudson commented on HDFS-7285: -- FAILURE: Integrated in Hadoop-Mapreduce-trunk #2408 (See [https://builds.apache.org/job/Hadoop-Mapreduce-trunk/2408/]) HDFS-8027. Erasure Coding: Update CHANGES-HDFS-7285.txt with branch (zhezhang: rev af8eaacdb40afba327b776e2b4f374be624289fc) * hadoop-hdfs-project/hadoop-hdfs/CHANGES-HDFS-EC-7285.txt HDFS-8027. Erasure Coding: Update CHANGES-HDFS-7285.txt with branch (zhezhang: rev 2c277802c1635837ec926862523be02e05e5649b) * hadoop-hdfs-project/hadoop-hdfs/CHANGES-HDFS-EC-7285.txt > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Fix For: 3.0.0 > > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-Consolidated-20150911.patch, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14938076#comment-14938076 ] Hudson commented on HDFS-7285: -- FAILURE: Integrated in Hadoop-Hdfs-trunk-Java8 #439 (See [https://builds.apache.org/job/Hadoop-Hdfs-trunk-Java8/439/]) HDFS-8027. Erasure Coding: Update CHANGES-HDFS-7285.txt with branch (zhezhang: rev af8eaacdb40afba327b776e2b4f374be624289fc) * hadoop-hdfs-project/hadoop-hdfs/CHANGES-HDFS-EC-7285.txt HDFS-8027. Erasure Coding: Update CHANGES-HDFS-7285.txt with branch (zhezhang: rev 2c277802c1635837ec926862523be02e05e5649b) * hadoop-hdfs-project/hadoop-hdfs/CHANGES-HDFS-EC-7285.txt > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Fix For: 3.0.0 > > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-Consolidated-20150911.patch, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14937857#comment-14937857 ] Hudson commented on HDFS-7285: -- FAILURE: Integrated in Hadoop-Yarn-trunk-Java8 #473 (See [https://builds.apache.org/job/Hadoop-Yarn-trunk-Java8/473/]) HDFS-8027. Erasure Coding: Update CHANGES-HDFS-7285.txt with branch (zhezhang: rev af8eaacdb40afba327b776e2b4f374be624289fc) * hadoop-hdfs-project/hadoop-hdfs/CHANGES-HDFS-EC-7285.txt HDFS-8027. Erasure Coding: Update CHANGES-HDFS-7285.txt with branch (zhezhang: rev 2c277802c1635837ec926862523be02e05e5649b) * hadoop-hdfs-project/hadoop-hdfs/CHANGES-HDFS-EC-7285.txt > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Fix For: 3.0.0 > > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-Consolidated-20150911.patch, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14937600#comment-14937600 ] Hudson commented on HDFS-7285: -- FAILURE: Integrated in Hadoop-Yarn-trunk #1203 (See [https://builds.apache.org/job/Hadoop-Yarn-trunk/1203/]) HDFS-8027. Erasure Coding: Update CHANGES-HDFS-7285.txt with branch (zhezhang: rev af8eaacdb40afba327b776e2b4f374be624289fc) * hadoop-hdfs-project/hadoop-hdfs/CHANGES-HDFS-EC-7285.txt HDFS-8027. Erasure Coding: Update CHANGES-HDFS-7285.txt with branch (zhezhang: rev 2c277802c1635837ec926862523be02e05e5649b) * hadoop-hdfs-project/hadoop-hdfs/CHANGES-HDFS-EC-7285.txt > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Fix For: 3.0.0 > > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-Consolidated-20150911.patch, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14937353#comment-14937353 ] Hudson commented on HDFS-7285: -- FAILURE: Integrated in Hadoop-trunk-Commit #8548 (See [https://builds.apache.org/job/Hadoop-trunk-Commit/8548/]) HDFS-8027. Erasure Coding: Update CHANGES-HDFS-7285.txt with branch (zhezhang: rev af8eaacdb40afba327b776e2b4f374be624289fc) * hadoop-hdfs-project/hadoop-hdfs/CHANGES-HDFS-EC-7285.txt HDFS-8027. Erasure Coding: Update CHANGES-HDFS-7285.txt with branch (zhezhang: rev 2c277802c1635837ec926862523be02e05e5649b) * hadoop-hdfs-project/hadoop-hdfs/CHANGES-HDFS-EC-7285.txt > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Fix For: 3.0.0 > > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-Consolidated-20150911.patch, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14937156#comment-14937156 ] Hudson commented on HDFS-7285: -- FAILURE: Integrated in Hadoop-Hdfs-trunk #2379 (See [https://builds.apache.org/job/Hadoop-Hdfs-trunk/2379/]) HDFS-8027. Erasure Coding: Update CHANGES-HDFS-7285.txt with branch (zhezhang: rev af8eaacdb40afba327b776e2b4f374be624289fc) * hadoop-hdfs-project/hadoop-hdfs/CHANGES-HDFS-EC-7285.txt HDFS-8027. Erasure Coding: Update CHANGES-HDFS-7285.txt with branch (zhezhang: rev 2c277802c1635837ec926862523be02e05e5649b) * hadoop-hdfs-project/hadoop-hdfs/CHANGES-HDFS-EC-7285.txt > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Fix For: 3.0.0 > > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-Consolidated-20150911.patch, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14936400#comment-14936400 ] Zhe Zhang commented on HDFS-7285: - The [vote | http://mail-archives.apache.org/mod_mbox/hadoop-hdfs-dev/201509.mbox/%3cCAAjoTXnx+79aGB=wV+NxQA9jy7KS9aw_b=19wmf9hbku0te...@mail.gmail.com%3e] on hdfs-dev and hadoop-dev mail lists to merge this feature branch to trunk has just passed. I will work with Andrew to merge the branch. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-Consolidated-20150911.patch, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14903897#comment-14903897 ] Zhe Zhang commented on HDFS-7285: - The Jenkins [job | https://builds.apache.org/job/Hadoop-HDFS-7285-Merge/] is not showing any new failed tests. I just updated the feature branch with latest trunk changes. It was a force push because HDFS-8920 was committed while I test the git merge result locally. So I just cherry-picked HDFS-8920; there was no conflict. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-Consolidated-20150911.patch, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14902180#comment-14902180 ] Zhe Zhang commented on HDFS-7285: - I just made another pass of the open subtasks. None of them looks like blockers for merging the feature branch to trunk. The nightly Jenkins [job | https://builds.apache.org/job/Hadoop-HDFS-7285-nightly/] is relatively stable. Most runs with massive failures are caused by stale dependencies (e.g. {{NoClassDefFoundError}} on {{ECSchema}}, after HDFS-8833 removed dependency on it). The only real test issue is {{TestWriteStripedFileWithFailure}}, which is flaky. But we already know that it depends on a series of open JIRAs on error handling in the write path (HDFS-8704, HDFS-9040). We under-estimated the required efforts on those JIRAs so decided to commit the test first. I think we can revert the patch at this phase and redo the change later. I also just made another pass of merging trunk changes into the feature branch locally, and pushed the result to my personal github [repo | https://github.com/zhe-thoughts/hadoop/tree/HDFS-7285-merge]. A Jenkins [job | https://builds.apache.org/job/Hadoop-HDFS-7285-Merge/] has been started. The main conflict was with HDFS-9111. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-Consolidated-20150911.patch, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14743823#comment-14743823 ] Zhe Zhang commented on HDFS-7285: - I just moved sub-tasks on write pipeline error handling to the follow-on umbrella, HDFS-8031. If we decide to change the target version we can move some of them back. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-Consolidated-20150911.patch, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14741643#comment-14741643 ] Zhe Zhang commented on HDFS-7285: - I agree with Jing that write pipeline error handling can be moved to follow-on. We should also think about whether to merge EC to branch-2, and if so, what should be the target version (current target version is 3.0.0). > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-Consolidated-20150911.patch, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14741620#comment-14741620 ] Jing Zhao commented on HDFS-7285: - Thanks Zhe! Also IMO the write pipeline error handling may not be a blocker for merging the feature branch to trunk (but may be for branch-2). > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-Consolidated-20150911.patch, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14723077#comment-14723077 ] Rui Li commented on HDFS-7285: -- Yep. We intend to create a throughput benchmark tool in HDFS-8968. HADOOP-11588 is something like a micro-benchmark, which benchmarks performance at an API level. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14723068#comment-14723068 ] Zhe Zhang commented on HDFS-7285: - Thanks Rui! I guess you are referring to HDFS-8968? > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14723066#comment-14723066 ] Zhe Zhang commented on HDFS-7285: - Thanks Walter! I updated the {{getBlockCollection}} in my personal github repo and the Jenkins job has been pretty stable so far. I'll incorporate your suggestions and update the main HDFS-7285 branch. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14723022#comment-14723022 ] Rui Li commented on HDFS-7285: -- [~brahmareddy] - Yeah I'm working on the two issues to support parallel read/write benchmarking. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14721869#comment-14721869 ] Brahma Reddy Battula commented on HDFS-7285: Parallel write/read meant for bench-marking the perf aspects and observing any concurrency issues.( different clients writing different files,read files..), I think, we can take care in (HDFS-8968 and HADOOP-11588). > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14720990#comment-14720990 ] Walter Su commented on HDFS-7285: - The merge result in HDFS-7285-20150827 LGTM. except that, 1. We have removed BlockInfo(BlockInfo from) constructor 2. Here It should be {{blocksmap.getStoredBlock(b)}}, and it could return {{null}}; {code} + public BlockCollection getBlockCollection(Block b) { //BlockManager.java - return blocksMap.getBlockCollection(b); ++BlockInfo storedBlock = getStoredBlock(b); ++return namesystem.getBlockCollection(storedBlock.getBlockCollectionId()); ++ } {code} I think it's better to not to call {{getStoredBlock}} inside {{getBlockCollection}}. It's wired to have another {{getBlockCollection}} which only used in {{SequentialBlockGroupIdGenerator}}. Can it just call blockManager.blocksMap.getStoredBlock(..) ? 3. wrong intent {code} - if (file.isStriped()) { //FSDirAttrOp.java - throw new UnsupportedActionException( - "Cannot set replication to a file with striped blocks"); - } ++ if (file.isStriped()) { ++throw new UnsupportedActionException( ++"Cannot set replication to a file with striped blocks"); ++ } {code} > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14718197#comment-14718197 ] GAO Rui commented on HDFS-7285: --- Thank you very much [~brahmareddy], [~zhz]. {quote} 1) snapshot feature 2) balancer feature {quote} may will be developed in future EC work, we could add these into the system test plan, and implement the test later. {quote} 4) parallel writes 5) parallel reads {quote} I think {{parallel reads}} means more than one client try to read the same EC file form HDFS, right? What is {{parallel writes}} refer to, in EC system testing? Could you explain the scenario? {quote} 1. Good points from Brahma Reddy Battula, I suggest that we also add HSM/mover tests to the list. 2. In reading tests we can distinguish stateful read and pread. Maybe we should test seek-and-read scenario too. 3. It seems each test scenario in the "Tips for EC Writing/Reading" section is systematically labeled. Will the labels be used to drive automatic testing? {quote} We can also add {{HSM/mover}} to the test plan, and implement it in future work. For the reading distinguish, we currently implement system test by using FSShell command in terminal, like {{CopyFromLocal}} and {{CopyToLocal}}. Can we set the client to read EC file in particular mechanism like stateful read and pread by terminal command? The labels in EC Writing/Reading tests were generated by test script during the test process, but it is also possible to drive automatic testing by the scenario labels vice versa. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, > HDFSErasureCodingSystemTestReport-20150826.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14717760#comment-14717760 ] Zhe Zhang commented on HDFS-7285: - I just finished another pass of "git merge" to bring HDFS-7285 branch in sync with trunk. There were quite a lot of conflicts so I pushed to my personal github [repo | https://github.com/zhe-thoughts/hadoop/tree/HDFS-7285-20150827] first. Also triggered a Jenkins [job | https://builds.apache.org/job/Hadoop-HDFS-7285-Merge/]. Once verified I will update the main HDFS-7285 branch. Meanwhile let me know if you see any issues. Most conflicts were with HDFS-8823 and recently committed subtasks of HDFS-6200, in case anyone wants to take a look at how conflicts were resolved. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14717085#comment-14717085 ] Zhe Zhang commented on HDFS-7285: - [~drankye] Thanks for raising the question. Since we have reached the conclusion to allow "git merge" in the workflow, I have merged trunk changes into the main HDFS-7285 branch. Going forward let's just use {{HDFS-7285}} branch for all commits. I have actually deleted the intermediate {{HDFS-7285-merge}} branch. [~vinayrpet] Do you think we should delete the {{HDFS-7285-REBASE}} branch too, given that the result has been compared against and incorporated into the current {{HDFS-7285}} branch? > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14715996#comment-14715996 ] Kai Zheng commented on HDFS-7285: - Thanks [~zhz] for the hard taking and I understand we may take some time to make the merging to trunk happen. Before that, I thought it would be good to keep moving. Think about so many follow-on issues listed in HDFS-8031 even for phase I, not to say we have another phase for non-striping erasure coding support. Having some branches for the effort, I'm wondering which branch better to use for the community to continuously work on and commit the follow-on tasks. Any thought? Thanks. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14715985#comment-14715985 ] Kai Zheng commented on HDFS-7285: - It's great to have the complete functional system tests for the striping support. Thanks guys! In our side, we're working on end to end performance tests while implementing new erasure coders and optimizing striping codes. Related tasks: * Implementing ISA-L coder (about 25X than the Java coder, HADOOP-11887, HADOOP-11540); * Implementing a new Java RS coder that’s both faster (6X than the existing Java coder) and compatible with ISA-L coder (HADOOP-12041). * Optimizing client striping input/output stream codes (HDFS-8668 and related), consolidating client stateful and positional reads (HDFS-8957 and related); * Performance benchmark test tools (HDFS-8968 and HADOOP-11588). We're running the perf tests while tuning and optimizing the codes, and will provide the report when it's ready. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14710770#comment-14710770 ] Zhe Zhang commented on HDFS-7285: - [~vinayrpet] Could you take a look at the results comparing the two rebase/merge efforts, and see if we should delete the {{HDFS-7285-REBASE}} branch? Thanks! > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14710239#comment-14710239 ] Zhe Zhang commented on HDFS-7285: - [~demongaorui] Thanks for the great addition to the test plan! The list of system functional tests looks good to me. A few questions that I think we can further discuss: # Good points from [~brahmareddy], I suggest that we also add HSM/mover tests to the list. # In reading tests we can distinguish stateful read and pread. Maybe we should test seek-and-read scenario too. # It seems each test scenario in the "Tips for EC Writing/Reading" section is systematically labeled. Will the labels be used to drive automatic testing? > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14710060#comment-14710060 ] Zhe Zhang commented on HDFS-7285: - Latest Jenkins [run | https://builds.apache.org/job/Hadoop-HDFS-7285-Merge/102/] is stable. I just updated the {{HDFS-7285}} branch as described above. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Compare-consolidated-20150824.diff, > Consolidated-20150707.patch, Consolidated-20150806.patch, > Consolidated-20150810.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14708955#comment-14708955 ] Brahma Reddy Battula commented on HDFS-7285: Thanks [~demongaorui] for uploading the HDFSErasureCodingSystemTestPlan-20150824.pdf. AFAIK following can be discussed and can be considered for testcases.. 1) snapshot feature 2) balancer feature 3) block placement policy based testing like, diferent racks etc.. I feel, better to consider following also.. 4) parallel writes 5) parallel reads > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, > Consolidated-20150806.patch, Consolidated-20150810.patch, ECAnalyzer.py, > ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > HDFSErasureCodingSystemTestPlan-20150824.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14705544#comment-14705544 ] Zhe Zhang commented on HDFS-7285: - Thanks Walter for reviewing the branch! I made the above changes and pushed to {{HDFS-7285-merge}} branch. Jenkins [reported | https://builds.apache.org/job/Hadoop-HDFS-7285-Merge/95/testReport/] 9 failures. They are all on hadoop-common, which shouldn't be affected by EC changes. They all pass locally. I'll compare the consolidated patches from this merge and Vinay's branch and report findings here. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, > Consolidated-20150806.patch, Consolidated-20150810.patch, ECAnalyzer.py, > ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14704910#comment-14704910 ] Walter Su commented on HDFS-7285: - Comments on HDFS-7285-merge branch. 1. Unused import. {code} +import com.google.common.base.Preconditions; //DFSInputStream.java {code} {code} +import org.apache.commons.logging.Log; //DFSOutputStream.java +import org.apache.commons.logging.LogFactory; {code} 3. Please replace it with {{stat.getErasureCodingPolicy()==null}} {code} +if(stat.getReplication() == 0) { //DFSOutputStream.java + throw new IOException("Not support appending to a striping layout file yet."); +} {code} 4. redundant space character. {code} -public abstract class BlockInfo extends Block //BlockInfo.java +public abstract class BlockInfo extends Block {code} 5. indent {code} - private LocatedBlock createLocatedBlock(final BlockInfo blk, final long pos) { //BlockManager.java + private LocatedBlock createLocatedBlock(final BlockInfo blk, final long pos + ) throws IOException { {code} 6. truncation is file level. Maybe assert !file.isStriped() is less confusing? And also {{createNewBlock(false)}} can be less confusing. {code} + BlockInfo oldBlock = file.getLastBlock(); //FSDirTruncateOp.java ++assert !oldBlock.isStriped(); ... ++ fsn.createNewBlock(file.isStriped()) : new Block( {code} 7. unnecessary change {code} -fsn.getBlockManager().setBlockToken(lBlk, //FSDirWriteFileOp.java -BlockTokenIdentifier.AccessMode.WRITE); +fsn.getFSDirectory().getBlockManager(). +setBlockToken(lBlk, BlockTokenIdentifier.AccessMode.WRITE); {code} 8. unnecessary change. (wiki says using [Sun's conventions|http://www.oracle.com/technetwork/java/javase/documentation/codeconventions-136091.html], which says "Break before an operator") {code} - if (blockInfo.getBlockCollection().getStoragePolicyID() //FSNamesystem.java - == lpPolicy.getId()) { // feature branch + BlockInfo blockInfo = getStoredBlock(b); - if (blockInfo.getBlockCollection().getStoragePolicyID() == lpPolicy.getId()) { // trunk ++ if (blockInfo.getBlockCollection().getStoragePolicyID() == ++ lpPolicy.getId()) { {code} Thanks [~zhz] for the great work! > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, > Consolidated-20150806.patch, Consolidated-20150810.patch, ECAnalyzer.py, > ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14704384#comment-14704384 ] Zhe Zhang commented on HDFS-7285: - I just finished step #1 and #2 above, and pushed the result of "git merge" to the {{HDFS-7285-merge}} branch. Jenkins [job | https://builds.apache.org/job/Hadoop-HDFS-7285-Merge/] has been triggered. Because HDFS-8801 requires major changes to the branch, this "git merge" was against HDFS-6407, which immediately precedes HDFS-8801 in trunk. [~jingzhao] has created a patch under HDFS-8909 which should be able to merge HDFS-8801 to the branch (I think we should keep it as a separate JIRA because it does more than just merging). > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, > Consolidated-20150806.patch, Consolidated-20150810.patch, ECAnalyzer.py, > ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14702000#comment-14702000 ] Zhe Zhang commented on HDFS-7285: - Given the feedbacks from the common-dev [discussion | http://mail-archives.apache.org/mod_mbox/hadoop-common-dev/201508.mbox/%3cCAGB5D2bPWeV2Hk+67=DamWpVfLTM6nkjb_wG3n4=wan890z...@mail.gmail.com%3e], I think we should use "git merge" to sync HDFS-7285 branch with trunk now. Please comment if you have any concern. I will do the following before updating the main HDFS-7285 branch: # Run "git merge" on current HDFS-7285 branch agains trunk, and add one merge commit to resolve all conflicts so far # Reuse the current HDFS-7285-merge branch to carry the merged result, and run Jenkins # Compare the merged result with the rebased result from [~vinayrpet] on HDFS-7285-REBASE branch If no major discrepancy is found, I will push the merged result to HDFS-7285 branch and delete the intermediate branches. I apologize for not foreseeing the complexity and size of the EC branch at the beginning of the project, and adjusting the git workflow accordingly. Hope we have cleared confusions through the above discussions. Let me know if you still have questions about the workflow. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, > Consolidated-20150806.patch, Consolidated-20150810.patch, ECAnalyzer.py, > ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14699916#comment-14699916 ] Zhe Zhang commented on HDFS-7285: - [~vinayrpet] Sure, let's solicit more feedback. Your rebase doesn't cause any additional failures: [Jenkins results | https://builds.apache.org/job/Hadoop-HDFS-7285-REBASE/]. Did you run Jenkins before posting the branch? Otherwise, a nice ace :) > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, > Consolidated-20150806.patch, Consolidated-20150810.patch, ECAnalyzer.py, > ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14699906#comment-14699906 ] Zhe Zhang commented on HDFS-7285: - Thanks Walter for the questions. bq. Will you rebase HDFS-7285 weekly after squashes it? Yes that's my plan. Actually since we are close to merging I plan to rebase more frequently. Currently I'm rebasing HDFS-7285-merge daily bq. Should we squash HDFS-8854 and HDFS-8833 as well? To make the future rebasing easier, also to try to avoid a second squash. [~andrew.wang] has started a discussion thread on common-dev regarding the rebase workflow. I'll wait until we reach a consensus there before squashing the 2 new big patches. bq. The commit message is inaccurate because HDFS-7285 is not finished yet. Good point. I'll reword it in next rebase. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, > Consolidated-20150806.patch, Consolidated-20150810.patch, ECAnalyzer.py, > ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14698914#comment-14698914 ] Walter Su commented on HDFS-7285: - bq. I'll also move the current HDFS-7285 branch as a backup. bq. Then we can decide about moving this as HDFS-7285 Branch. Great. I have some question [~zhz]. 1. Will you rebase HDFS-7285 weekly after squashes it? 2. Should we squash HDFS-8854 and HDFS-8833 as well? To make the future rebasing easier, also to try to avoid a second squash. 3. The commit message is inaccurate because HDFS-7285 is not finished yet. {noformat} commit ecf3634830fdbf228c056e3d4ae77ae0def17683 ... HDFS-7285. Erasure Coding Support inside HDFS. {noformat} > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, > Consolidated-20150806.patch, Consolidated-20150810.patch, ECAnalyzer.py, > ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14698093#comment-14698093 ] Vinayakumar B commented on HDFS-7285: - Thanks for starting jenkins job. I prefer to wait for sometime for others also to take a look at the rebase. Then we can decide about moving this as HDFS-7285 Branch. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, > Consolidated-20150806.patch, Consolidated-20150810.patch, ECAnalyzer.py, > ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14697956#comment-14697956 ] Zhe Zhang commented on HDFS-7285: - Many thanks to Vinay for the great effort! I just cherry-picked the 2 new commits (HDFS-8854 and HDFS-8220) to the branch. Also created a Jenkins [job | https://builds.apache.org/job/Hadoop-HDFS-7285-nightly/]. I'll also compare {{HDFS-7285-REBASE}} with the consolidated patch. After that and verifying Jenkins results, I'll push it as {{HDFS-7285}} so we can better proceed with pending subtasks. I'll also move the current {{HDFS-7285}} branch as a backup, in case we want to reconcile differences in individual commits. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, > Consolidated-20150806.patch, Consolidated-20150810.patch, ECAnalyzer.py, > ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14696880#comment-14696880 ] Vinayakumar B commented on HDFS-7285: - I have tried to rebase current {{HDFS-7285}} branch against the current trunk using {{git rebase trunk}}. It was not smooth as expected. Since I did not wanted to push the rebase directly onto {{HDFS-7285}}, created one more branch {{HDFS-7285-REBASE}}. This branch is just for reference purpose. The advantage of this is, it retained all the commits along with message,date and author details, even after resolving conflicts. I skipped one commit purposefully HDFS-8787 to be in sync with trunk. it was just rename of files. other than this, no other commits got squashed. There were 192 commits to be rebased against trunk, including the intermediate merge conflict resolved commits. Since I couldnt edit each and every commit to resolve compilation errors after each commit, resolved remaining compilation errors at the end, with one more commit. If anyone wants to verify, please checkout the branch HDFS-7285-REBASE. and can compare against the Consolidated patch. Since this is only for trying to check the possibility of rebase, I am not saying this should be considered as final branch. If everyone feels good to go like this approach, I could do some more detailed rebase next week, ( may be verify the compilation after each commit ? Not sure whether its possible to stop for each commit rebase) -Thanks > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, > Consolidated-20150806.patch, Consolidated-20150810.patch, ECAnalyzer.py, > ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14681311#comment-14681311 ] Zhe Zhang commented on HDFS-7285: - As explained below, I think each subtask should *at least* be committed to HDFS-7285-merge}}. For HDFS-8854, since we already have 2 versions, I think we should commit to both. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, > Consolidated-20150806.patch, Consolidated-20150810.patch, ECAnalyzer.py, > ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14681307#comment-14681307 ] Zhe Zhang commented on HDFS-7285: - [~wheat9] [~jingzhao] [~walter.k.su] Thanks for the discussion. bq. at the very least you will need consensuses with other contributors anyway. Absolutely. I created the {{HDFS-7285-merge}} branch exactly because we haven't reached a consensus on whether / how to squash the commits in the current HDFS-7285 branch when rebased against trunk. Meanwhile, we do need to commit patches of the pending subtasks. Committing to the current HDFS-7285 branch will cause additional rebasing efforts. bq. The new feature branch Zhe proposed here is for merging the changes from trunk to EC dev. Still doing git rebase may be too hard now. Thanks Jing for helping explaining it. Yes the new branch is just *HDFS-7285 rebased against trunk* (or from another angle, merging trunk changes to EC). We can continue discussing how to split the git history, but the end result (the {{HEAD}}) won't be affected. bq. Can we just sync HDFS-7285 with trunk? {{HDFS-7285-merge}} is actually already synced with trunk. If we all agree, it's actually clearer if we push it as {{HDFS-7285}} and move the current {{HDFS-7285}} as a backup branch. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, > Consolidated-20150806.patch, Consolidated-20150810.patch, ECAnalyzer.py, > ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14681079#comment-14681079 ] Walter Su commented on HDFS-7285: - For synchronizing branch with trunk, we can use 'git merge'. For merging branch to trunk, we can use 14 squashed sub-patches the way [~zhz] did. Is it weird that one commit has a huge commit message? We commit one sub-patch, and keep many jira numbers and contributors as a whole commit message for this one commit log. In this way, we can avoid the painful process of rebasing every out-of-date commit, and also can keep the contributor infos as [~szetszwo] suggested. I don't know if it's weird. At least it's an option. bq. So if you are working on one of the open subtasks, please kindly target the patch on HDFS-7285-merge. Ideally we should also commit to HDFS-7285 branch So after HDFS-8854. An open subtask has to prepare 2 patches. a 'ECPolicy' version for HDFS-7285-merge, and a 'ECSchema' version for HDFS-7285? Or we commit HDFS-8854 to both of them first? Can we just sync HDFS-7285 with trunk? > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, > Consolidated-20150806.patch, Consolidated-20150810.patch, ECAnalyzer.py, > ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14681006#comment-14681006 ] Jing Zhao commented on HDFS-7285: - Looks like the current situation is little bit confusing :) The new feature branch Zhe proposed here is for merging the changes from trunk to EC dev. Still doing git rebase may be too hard now. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, > Consolidated-20150806.patch, Consolidated-20150810.patch, ECAnalyzer.py, > ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14680995#comment-14680995 ] Haohui Mai commented on HDFS-7285: -- bq. The new branch was created simply because we haven't reached an agreement on the rebase workflow. I don't quite get what you are trying to achieve here. If the whole point of the branch, however, is about merging it back to the mainline, at the very least you will need consensuses with other contributors anyway. No? > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, > Consolidated-20150806.patch, Consolidated-20150810.patch, ECAnalyzer.py, > ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14680972#comment-14680972 ] Zhe Zhang commented on HDFS-7285: - [~szetszwo]: Thanks for the question. I posted my planned rebase workflow above: [link 1| https://issues.apache.org/jira/browse/HDFS-7285?focusedCommentId=14593827&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-14593827], [link 2 | https://issues.apache.org/jira/browse/HDFS-7285?focusedCommentId=14600282&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-14600282], and a few other places. Based on that plan I already finished rebasing the branch with 14 squashed sub-patches. Then before pushing to HDFS-7285 branch we started revisiting HDFS-8499 and the rebase workflow. The new branch was created simply because we haven't reached an agreement on the rebase workflow. It enables us to commit pending patches regardless of the final conclusion on rebase workflow. Please let me know if that answers your question. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, > Consolidated-20150806.patch, Consolidated-20150810.patch, ECAnalyzer.py, > ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14680887#comment-14680887 ] Andrew Wang commented on HDFS-7285: --- [~szetszwo]: * If you have issues with the rebase workflow, let's take it to common-dev. This topic applies beyond the scope of erasure coding. * Regarding the refactors, you are talking about an SVN-style merge workflow rather than a git-style rebase workflow. In the case of BlockInfo, it went through a few mutations on the EC branch before arriving at the current state. This is because the understanding of BlockInfo+EC evolved over the course of development. It is prudent to wait until then to do the same refactoring in trunk, to avoid unnecessary churn on trunk. * Regarding credit, the original contributor gets credit on the JIRA targeted for the branch, yes? JIRA assignee is how we credit contributors, and if there are multiple contributors, a JIRA comment saying as much. If you have examples, let's correct assignees or add comments to make sure that all contributors are being properly credited. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, > Consolidated-20150806.patch, Consolidated-20150810.patch, ECAnalyzer.py, > ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14680868#comment-14680868 ] Tsz Wo Nicholas Sze commented on HDFS-7285: --- > Zhe Zhang, could you describe the git rebase workflow we are using? ... [~zhz], could you describe the workflow you are using? We have a lot of confusion since there are so many merge patches; we now have a new branch; and we stopped rebasing the HDFS-7285 at some point. It seems quite arbitrary. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, > Consolidated-20150806.patch, Consolidated-20150810.patch, ECAnalyzer.py, > ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14680864#comment-14680864 ] Tsz Wo Nicholas Sze commented on HDFS-7285: --- > A rebase workflow gets very difficult without the ability to squash patches, > since you would otherwise spend a lot of time fixing conflicts in > intermediate commits that don't even show up in HEAD. We have 180 some > commits in the EC branch now, and is definitely at the "very difficult" stage > of rebasing. If it is true, choosing the git rebase workflow seems a mistake. We probably should switch to git merge workflow then. > Moving refactors from branches to trunk is standard practice we've done many > times before and is something I recommended we do here too. The standard practice is to have a refactor patch committed to trunk first and then use git to merge to the branch, not the other way around, not using separated patches. Committing similar code to different branches by different patches makes the merge hard since the branch and trunk are going to have a lot of conflicts as a result of such practice. Moreover, if separated patches are needed, we usually ask the original contributors to contribute a merge patch for for committing patches to different branches so that the original contributors get the credits but not a different contributor. An original contributor A did the hard work to come up patches which were only committed to the development branch. Another contributor B did the easy work to post similar patches for committing to trunk. After the development branch got merged to trunk, no one would care about the development branch and all the contributions in trunk are associated to B but not the original contributor A. Although the contributor B may not have the intention, it does look like that B is stealing the credits from A. Do you think that it is a problem, [~andrew.wang]? > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, > Consolidated-20150806.patch, Consolidated-20150810.patch, ECAnalyzer.py, > ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14680785#comment-14680785 ] Hadoop QA commented on HDFS-7285: - \\ \\ | (x) *{color:red}-1 overall{color}* | \\ \\ || Vote || Subsystem || Runtime || Comment || | {color:red}-1{color} | patch | 0m 1s | The patch command could not apply the patch during dryrun. | \\ \\ || Subsystem || Report/Notes || | Patch URL | http://issues.apache.org/jira/secure/attachment/12749674/Consolidated-20150810.patch | | Optional Tests | javadoc javac unit findbugs checkstyle shellcheck | | git revision | HDFS-7285 / fbf7e81 | | Console output | https://builds.apache.org/job/PreCommit-HDFS-Build/11955/console | This message was automatically generated. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, > Consolidated-20150806.patch, Consolidated-20150810.patch, ECAnalyzer.py, > ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14680764#comment-14680764 ] Zhe Zhang commented on HDFS-7285: - To better proceed with the open subtasks, I just created a branch {{HDFS-7285-merge}}. It applies the rebased consolidated patch ({{Consolidated-20150810.patch}}) on top of trunk; I will rebase the branch daily. The branch basically provides an up-to-date {{HEAD}} for pending patches. After we finalize the above rebase workflow discussion we can always update the branch without changing the current {{HEAD}}. So if you are working on one of the open subtasks, please kindly target the patch on {{HDFS-7285-merge}}. Ideally we should also commit to {{HDFS-7285}} branch in case we want to keep the branch to reserve development history. If you have a WiP patch that's non-trivial to rebase I'm very happy to help. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, > Consolidated-20150806.patch, Consolidated-20150810.patch, ECAnalyzer.py, > ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14662742#comment-14662742 ] Andrew Wang commented on HDFS-7285: --- [~szetszwo] I never said that, what I did say was the following: * A rebase workflow gets very difficult without the ability to squash patches, since you would otherwise spend a lot of time fixing conflicts in intermediate commits that don't even show up in HEAD. We have 180 some commits in the EC branch now, and is definitely at the "very difficult" stage of rebasing. * Moving refactors from branches to trunk is standard practice we've done many times before and is something I recommended we do here too. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, > Consolidated-20150806.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14662366#comment-14662366 ] Tsz Wo Nicholas Sze commented on HDFS-7285: --- [~zhz], could you describe the git rebase workflow we are using? In a separated discussion, [~andrew.wang] claimed that our git rebase workflow requires committing some patches to trunk before merging the branch. I wonder if you feel the same way. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, > Consolidated-20150806.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14661421#comment-14661421 ] Hadoop QA commented on HDFS-7285: - \\ \\ | (x) *{color:red}-1 overall{color}* | \\ \\ || Vote || Subsystem || Runtime || Comment || | {color:red}-1{color} | patch | 0m 1s | The patch command could not apply the patch during dryrun. | \\ \\ || Subsystem || Report/Notes || | Patch URL | http://issues.apache.org/jira/secure/attachment/12749207/Consolidated-20150806.patch | | Optional Tests | javadoc javac unit findbugs checkstyle shellcheck | | git revision | HDFS-7285 / 8799363 | | Console output | https://builds.apache.org/job/PreCommit-HDFS-Build/11933/console | This message was automatically generated. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, > Consolidated-20150806.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14618936#comment-14618936 ] Zhe Zhang commented on HDFS-7285: - Thanks Jing for the helpful comments! I've created HDFS-8728 focusing on the {{BlockInfo}} and {{INodeFile}} changes. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
Auto-Re: [jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
您的邮件已收到!谢谢!
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14617804#comment-14617804 ] Hadoop QA commented on HDFS-7285: - \\ \\ | (x) *{color:red}-1 overall{color}* | \\ \\ || Vote || Subsystem || Runtime || Comment || | {color:red}-1{color} | patch | 0m 1s | The patch command could not apply the patch during dryrun. | \\ \\ || Subsystem || Report/Notes || | Patch URL | http://issues.apache.org/jira/secure/attachment/12744101/Consolidated-20150707.patch | | Optional Tests | javadoc javac unit findbugs checkstyle shellcheck | | git revision | HDFS-7285 / 2c494a8 | | Console output | https://builds.apache.org/job/PreCommit-HDFS-Build/11613/console | This message was automatically generated. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14617695#comment-14617695 ] Jing Zhao commented on HDFS-7285: - Similar for the changes on INodeFile. In general I suggest let's try our best to separate the code change from the merging. We're now merging the changes from trunk back to HDFS-7285, thus it will be much faster and safer to make the merged code same with the current HDFS-7285 branch. For all the other changes we can do them later as subtasks under HDFS-7285. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14617684#comment-14617684 ] Jing Zhao commented on HDFS-7285: - Thanks for the work, Zhe! So for the BlockInfoStriped related work, I suggest opening a separate jira under HDFS-7285 and post the main changes (from the current HDFS-7285 implementation) there. This can simplify the review. Also is it possible that we can defer this change after merging the remaining code? > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: Consolidated-20150707.patch, ECAnalyzer.py, ECParser.py, > HDFS-7285-initial-PoC.patch, HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-7285-merge-consolidated.trunk.03.patch, > HDFS-7285-merge-consolidated.trunk.04.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14611337#comment-14611337 ] Hadoop QA commented on HDFS-7285: - \\ \\ | (x) *{color:red}-1 overall{color}* | \\ \\ || Vote || Subsystem || Runtime || Comment || | {color:red}-1{color} | patch | 0m 1s | The patch command could not apply the patch during dryrun. | \\ \\ || Subsystem || Report/Notes || | Patch URL | http://issues.apache.org/jira/secure/attachment/12743179/HDFS-EC-merge-consolidated-01.patch | | Optional Tests | javadoc javac unit findbugs checkstyle shellcheck | | git revision | HDFS-7285 / 0b7af27 | | Console output | https://builds.apache.org/job/PreCommit-HDFS-Build/11569/console | This message was automatically generated. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: ECAnalyzer.py, ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-EC-merge-consolidated-01.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
Auto-Re: [jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
您的邮件已收到!谢谢!
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14610826#comment-14610826 ] Hadoop QA commented on HDFS-7285: - \\ \\ | (x) *{color:red}-1 overall{color}* | \\ \\ || Vote || Subsystem || Runtime || Comment || | {color:red}-1{color} | patch | 0m 1s | The patch command could not apply the patch during dryrun. | \\ \\ || Subsystem || Report/Notes || | Patch URL | http://issues.apache.org/jira/secure/attachment/12743121/HDFS-7285-merge-consolidated-trunk-01.patch | | Optional Tests | javadoc javac unit findbugs checkstyle shellcheck | | git revision | HDFS-7285 / 0b7af27 | | Console output | https://builds.apache.org/job/PreCommit-HDFS-Build/11565/console | This message was automatically generated. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: ECAnalyzer.py, ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-bistriped.patch, > HDFSErasureCodingDesign-20141028.pdf, HDFSErasureCodingDesign-20141217.pdf, > HDFSErasureCodingDesign-20150204.pdf, HDFSErasureCodingDesign-20150206.pdf, > HDFSErasureCodingPhaseITestPlan.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
Auto-Re: [jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
您的邮件已收到!谢谢!
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14610710#comment-14610710 ] Vinayakumar B commented on HDFS-7285: - Sure. Actually, I created this for this purpose only :). So that nothing should miss. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: ECAnalyzer.py, ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-bistriped.patch, > HDFSErasureCodingDesign-20141028.pdf, HDFSErasureCodingDesign-20141217.pdf, > HDFSErasureCodingDesign-20150204.pdf, HDFSErasureCodingDesign-20150206.pdf, > HDFSErasureCodingPhaseITestPlan.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
Auto-Re: [jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
您的邮件已收到!谢谢!
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14610696#comment-14610696 ] Zhe Zhang commented on HDFS-7285: - Thanks Vinay for the great effort! When I finish my current pass to divide the patch, we can compare the 2 consolidated patches and resolve all conflicts. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: ECAnalyzer.py, ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-bistriped.patch, > HDFSErasureCodingDesign-20141028.pdf, HDFSErasureCodingDesign-20141217.pdf, > HDFSErasureCodingDesign-20150204.pdf, HDFSErasureCodingDesign-20150206.pdf, > HDFSErasureCodingPhaseITestPlan.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
Auto-Re: [jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
您的邮件已收到!谢谢!
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14610688#comment-14610688 ] Zhe Zhang commented on HDFS-7285: - If Jenkins still tries to apply against HDFS-7285, we can rename the patch to HDFS-EC-xxx > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: ECAnalyzer.py, ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, > HDFS-7285-merge-consolidated-trunk-01.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-bistriped.patch, > HDFSErasureCodingDesign-20141028.pdf, HDFSErasureCodingDesign-20141217.pdf, > HDFSErasureCodingDesign-20150204.pdf, HDFSErasureCodingDesign-20150206.pdf, > HDFSErasureCodingPhaseITestPlan.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
Auto-Re: [jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
您的邮件已收到!谢谢!
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14610247#comment-14610247 ] Hadoop QA commented on HDFS-7285: - \\ \\ | (x) *{color:red}-1 overall{color}* | \\ \\ || Vote || Subsystem || Runtime || Comment || | {color:red}-1{color} | patch | 0m 1s | The patch command could not apply the patch during dryrun. | \\ \\ || Subsystem || Report/Notes || | Patch URL | http://issues.apache.org/jira/secure/attachment/12743065/HDFS-7285-merge-consolidated-01.patch | | Optional Tests | javadoc javac unit findbugs checkstyle shellcheck | | git revision | HDFS-7285 / 0b7af27 | | Console output | https://builds.apache.org/job/PreCommit-HDFS-Build/11561/console | This message was automatically generated. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: ECAnalyzer.py, ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-7285-merge-consolidated-01.patch, HDFS-EC-Merge-PoC-20150624.patch, > HDFS-bistriped.patch, HDFSErasureCodingDesign-20141028.pdf, > HDFSErasureCodingDesign-20141217.pdf, HDFSErasureCodingDesign-20150204.pdf, > HDFSErasureCodingDesign-20150206.pdf, HDFSErasureCodingPhaseITestPlan.pdf, > fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14609517#comment-14609517 ] Kai Zheng commented on HDFS-7285: - There're some changes in the patch 1 (for codec) that should be placed elsewhere. In {{FSOutputSummer}}, better for 9 (datanode support) {code} + protected DataChecksum getDataChecksum() { +return sum; + } {code} In {{FsShell}}, better for 7 (client side support) {code} + protected String getUsagePrefix() { +return usagePrefix; + } {code} I will continue to look at some other parts. And by the way, I'm going to move left codec issues planned for HDFS-7285 to the follow-on issue since I don't want to interrupt this pre-merge effort. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: ECAnalyzer.py, ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-bistriped.patch, > HDFSErasureCodingDesign-20141028.pdf, HDFSErasureCodingDesign-20141217.pdf, > HDFSErasureCodingDesign-20150204.pdf, HDFSErasureCodingDesign-20150206.pdf, > HDFSErasureCodingPhaseITestPlan.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
Auto-Re: [jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
您的邮件已收到!谢谢!
Auto-Re: [jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
您的邮件已收到!谢谢!
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14609070#comment-14609070 ] Zhe Zhang commented on HDFS-7285: - Thanks Kai! Feel free to claim the component you'd like to look at. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: ECAnalyzer.py, ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-bistriped.patch, > HDFSErasureCodingDesign-20141028.pdf, HDFSErasureCodingDesign-20141217.pdf, > HDFSErasureCodingDesign-20150204.pdf, HDFSErasureCodingDesign-20150206.pdf, > HDFSErasureCodingPhaseITestPlan.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
Auto-Re: [jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
您的邮件已收到!谢谢!
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14606741#comment-14606741 ] Kai Zheng commented on HDFS-7285: - Thanks [~zhz] for the great work! I will have some time to look at some parts I'm familiar with. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: ECAnalyzer.py, ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-bistriped.patch, > HDFSErasureCodingDesign-20141028.pdf, HDFSErasureCodingDesign-20141217.pdf, > HDFSErasureCodingDesign-20150204.pdf, HDFSErasureCodingDesign-20150206.pdf, > HDFSErasureCodingPhaseITestPlan.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
Auto-Re: [jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
您的邮件已收到!谢谢!
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14606639#comment-14606639 ] Zhe Zhang commented on HDFS-7285: - I just finished another pass of the consolidated patch. The github [repo | https://github.com/zhe-thoughts/hadoop/tree/HDFS-EC-Merge] has the latest code. I divided the consolidated patch into 13 sub-patches. Please let me know if the list looks reasonable to you. {code} 13. Balancer and mover support for striped block groups 12. Support striped block groups in fsimage and edit logs 11. Change fsck to support EC files 10. Add striped block support in INodeFile. 9. Datanode support 8. Distribute recovery work for striped blocks to DataNode. 7. Client side support 6. Create LocatedStripedBlock abstraction to represent striped block groups. 5. BlockPlacementPolicies for erasure coding 4. Allocate and manage striped blocks in NameNode blockmanagement module. 3. Extend BlockInfo to handle striped block groups. 2. Support Erasure Coding Zones. 1. HADOOP-COMMON side support for codec calculations. {code} I still haven't added tests to the sub-patches. Also the {{INodeFile}} implementation is still being discussed. They'll be addressed in the next pass. In the next pass, to ensure we capture all latest branch changes, I will do the following: # Take a latest consolidated patch from HDFS-7285 # Examine each file (if necessary, each diff in a file) in the patch and fit them into one of the 13 sub-patches (or discard as pre-merged changes). The consolidated patch has roughly 25k lines of code. So any volunteered help is much appreciated :) > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: ECAnalyzer.py, ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-bistriped.patch, > HDFSErasureCodingDesign-20141028.pdf, HDFSErasureCodingDesign-20141217.pdf, > HDFSErasureCodingDesign-20150204.pdf, HDFSErasureCodingDesign-20150206.pdf, > HDFSErasureCodingPhaseITestPlan.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
Auto-Re: [jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
您的邮件已收到!谢谢!
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14602463#comment-14602463 ] Zhe Zhang commented on HDFS-7285: - Thanks Walter! Will address in the next rev. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: ECAnalyzer.py, ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-bistriped.patch, > HDFSErasureCodingDesign-20141028.pdf, HDFSErasureCodingDesign-20141217.pdf, > HDFSErasureCodingDesign-20150204.pdf, HDFSErasureCodingDesign-20150206.pdf, > HDFSErasureCodingPhaseITestPlan.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
Auto-Re: [jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
您的邮件已收到!谢谢!
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14601018#comment-14601018 ] Walter Su commented on HDFS-7285: - Comments on github branch(PoC-20150624.patch): 1. You missed merge {{INodeFile#getLastBlock}} and {{numBlocks}} to github branch. 2. I run test, and saw NPE. It'll fix after HDFS-8653 merges into trunk. 3. Please remove this line because NPE when lastBlock==null. {code} 2437 Preconditions.checkState(!lastBlock.isStriped()); //FSNamesystem#appendFileInternal() {code} > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: ECAnalyzer.py, ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-bistriped.patch, > HDFSErasureCodingDesign-20141028.pdf, HDFSErasureCodingDesign-20141217.pdf, > HDFSErasureCodingDesign-20150204.pdf, HDFSErasureCodingDesign-20150206.pdf, > HDFSErasureCodingPhaseITestPlan.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
Auto-Re: [jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
您的邮件已收到!谢谢!
[jira] [Commented] (HDFS-7285) Erasure Coding Support inside HDFS
[ https://issues.apache.org/jira/browse/HDFS-7285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14600792#comment-14600792 ] Walter Su commented on HDFS-7285: - bq. As Jing and myself commented under HDFS-8058, the downside is weaker type safety. Hmm. I was worrying mixed types of UC/non-UC. You are worrying mixed types of striped/contiguous. Isn't type check in {{setBlock}} enough? I don't know. > Erasure Coding Support inside HDFS > -- > > Key: HDFS-7285 > URL: https://issues.apache.org/jira/browse/HDFS-7285 > Project: Hadoop HDFS > Issue Type: New Feature >Reporter: Weihua Jiang >Assignee: Zhe Zhang > Attachments: ECAnalyzer.py, ECParser.py, HDFS-7285-initial-PoC.patch, > HDFS-EC-Merge-PoC-20150624.patch, HDFS-bistriped.patch, > HDFSErasureCodingDesign-20141028.pdf, HDFSErasureCodingDesign-20141217.pdf, > HDFSErasureCodingDesign-20150204.pdf, HDFSErasureCodingDesign-20150206.pdf, > HDFSErasureCodingPhaseITestPlan.pdf, fsimage-analysis-20150105.pdf > > > Erasure Coding (EC) can greatly reduce the storage overhead without sacrifice > of data reliability, comparing to the existing HDFS 3-replica approach. For > example, if we use a 10+4 Reed Solomon coding, we can allow loss of 4 blocks, > with storage overhead only being 40%. This makes EC a quite attractive > alternative for big data storage, particularly for cold data. > Facebook had a related open source project called HDFS-RAID. It used to be > one of the contribute packages in HDFS but had been removed since Hadoop 2.0 > for maintain reason. The drawbacks are: 1) it is on top of HDFS and depends > on MapReduce to do encoding and decoding tasks; 2) it can only be used for > cold files that are intended not to be appended anymore; 3) the pure Java EC > coding implementation is extremely slow in practical use. Due to these, it > might not be a good idea to just bring HDFS-RAID back. > We (Intel and Cloudera) are working on a design to build EC into HDFS that > gets rid of any external dependencies, makes it self-contained and > independently maintained. This design lays the EC feature on the storage type > support and considers compatible with existing HDFS features like caching, > snapshot, encryption, high availability and etc. This design will also > support different EC coding schemes, implementations and policies for > different deployment scenarios. By utilizing advanced libraries (e.g. Intel > ISA-L library), an implementation can greatly improve the performance of EC > encoding/decoding and makes the EC solution even more attractive. We will > post the design document soon. -- This message was sent by Atlassian JIRA (v6.3.4#6332)