We found that a full java heap PermSpace caused the datanode process to be slow. With the high replication factor, the delay seems to build up with the number of datanodes. Increasing the PermSpace improved performance significantly and made impala a lot faster as well. Any other java settings that someone recommends having a look at to improve performance?
Cheers, Laurens -----Original Message----- From: Laurens Bronwasser Sent: Tuesday, September 02, 2014 03:47 PM W. Europe Standard Time To: user@hadoop.apache.org Cc: Julien Lehuen; Tyler McDougall Subject: Re: Replication factor affecting write performance Thanks Stanley for responding. The size of the cluster B is 12 data nodes. Cluster A has 18 (slower, older) data nodes. The topology is that all data nodes and name nodes are under one switch. The clusters are not sharing a switch though, the cluster are in different data centers. Perhaps this helps, below is a trace of the combined logs of all data nodes, after writing one block of 500MB with replication factor 12, on Cluster B. Note that the logs are combined and sorted on millisecond precision, certain lines may have happened in a different order: 2014-09-02 15:18:41,135 INFO org.apache.hadoop.hdfs.server.datanode.DataNode: Receiving BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587 src: /10.141.0.87:55341 dest: /10.141.0.159:50010 hinterlands.trading.imc.intra 2014-09-02 15:18:41,137 INFO org.apache.hadoop.hdfs.server.datanode.DataNode: Receiving BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587 src: /10.141.0.159:46759 dest: /10.141.0.168:50010 redridge.trading.imc.intra 2014-09-02 15:18:41,138 INFO org.apache.hadoop.hdfs.server.datanode.DataNode: Receiving BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587 src: /10.141.0.168:37742 dest: /10.141.0.157:50010 eversong.trading.imc.intra 2014-09-02 15:18:41,139 INFO org.apache.hadoop.hdfs.server.datanode.DataNode: Receiving BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587 src: /10.141.0.157:36874 dest: /10.141.0.156:50010 elwynn.trading.imc.intra 2014-09-02 15:18:41,140 INFO org.apache.hadoop.hdfs.server.datanode.DataNode: Receiving BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587 src: /10.141.0.156:43356 dest: /10.141.0.170:50010 searing.trading.imc.intra 2014-09-02 15:18:41,141 INFO org.apache.hadoop.hdfs.server.datanode.DataNode: Receiving BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587 src: /10.141.0.170:35201 dest: /10.141.0.155:50010 duskwood.trading.imc.intra 2014-09-02 15:18:41,142 INFO org.apache.hadoop.hdfs.server.datanode.DataNode: Receiving BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587 src: /10.141.0.155:52119 dest: /10.141.0.174:50010 tirisfal.trading.imc.intra 2014-09-02 15:18:41,143 INFO org.apache.hadoop.hdfs.server.datanode.DataNode: Receiving BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587 src: /10.141.0.174:45775 dest: /10.141.0.173:50010 sorrows.trading.imc.intra 2014-09-02 15:18:41,144 INFO org.apache.hadoop.hdfs.server.datanode.DataNode: Receiving BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587 src: /10.141.0.173:55624 dest: /10.141.0.171:50010 shimmering.trading.imc.intra 2014-09-02 15:18:41,145 INFO org.apache.hadoop.hdfs.server.datanode.DataNode: Receiving BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587 src: /10.141.0.171:45532 dest: /10.141.0.172:50010 silverpine.trading.imc.intra 2014-09-02 15:18:41,146 INFO org.apache.hadoop.hdfs.server.datanode.DataNode: Receiving BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587 src: /10.141.0.172:35460 dest: /10.141.0.158:50010 gilneas.trading.imc.intra 2014-09-02 15:18:41,147 INFO org.apache.hadoop.hdfs.server.datanode.DataNode: Receiving BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587 src: /10.141.0.158:59490 dest: /10.141.0.169:50010 scarlet.trading.imc.intra 2014-09-02 15:19:11,506 INFO org.apache.hadoop.hdfs.server.datanode.DataNode.clienttrace: src: /10.141.0.158:59490, dest: /10.141.0.169:50010, bytes: 536000000, op: HDFS_WRITE, cliID: DFSClient_NONMAPREDUCE_1486044046_1, offset: 0, srvID: f18f77e8-db9a-4e91-9160-401ae4b2f15b, blockid: BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587, duration: 30357265647 scarlet.trading.imc.intra 2014-09-02 15:19:11,506 INFO org.apache.hadoop.hdfs.server.datanode.DataNode.clienttrace: src: /10.141.0.172:35460, dest: /10.141.0.158:50010, bytes: 536000000, op: HDFS_WRITE, cliID: DFSClient_NONMAPREDUCE_1486044046_1, offset: 0, srvID: 66b3bcb2-36a8-4684-878f-2b90cec9d558, blockid: BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587, duration: 30357606668 gilneas.trading.imc.intra 2014-09-02 15:19:11,506 INFO org.apache.hadoop.hdfs.server.datanode.DataNode: PacketResponder: BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587, type=HAS_DOWNSTREAM_IN_PIPELINE terminating gilneas.trading.imc.intra 2014-09-02 15:19:11,506 INFO org.apache.hadoop.hdfs.server.datanode.DataNode: PacketResponder: BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587, type=LAST_IN_PIPELINE, downstreams=0:[] terminating scarlet.trading.imc.intra 2014-09-02 15:19:11,507 INFO org.apache.hadoop.hdfs.server.datanode.DataNode.clienttrace: src: /10.141.0.171:45532, dest: /10.141.0.172:50010, bytes: 536000000, op: HDFS_WRITE, cliID: DFSClient_NONMAPREDUCE_1486044046_1, offset: 0, srvID: 3d587891-c434-4a45-acb1-e41122ab71f9, blockid: BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587, duration: 30357881405 silverpine.trading.imc.intra 2014-09-02 15:19:11,507 INFO org.apache.hadoop.hdfs.server.datanode.DataNode.clienttrace: src: /10.141.0.173:55624, dest: /10.141.0.171:50010, bytes: 536000000, op: HDFS_WRITE, cliID: DFSClient_NONMAPREDUCE_1486044046_1, offset: 0, srvID: 3b0fcf75-78d9-4bcd-b5e0-941a8dd40da8, blockid: BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587, duration: 30358283146 shimmering.trading.imc.intra 2014-09-02 15:19:11,507 INFO org.apache.hadoop.hdfs.server.datanode.DataNode.clienttrace: src: /10.141.0.174:45775, dest: /10.141.0.173:50010, bytes: 536000000, op: HDFS_WRITE, cliID: DFSClient_NONMAPREDUCE_1486044046_1, offset: 0, srvID: 60d4ad85-90d0-42e1-958e-b73ff55a4d2c, blockid: BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587, duration: 30358543752 sorrows.trading.imc.intra 2014-09-02 15:19:11,507 INFO org.apache.hadoop.hdfs.server.datanode.DataNode: PacketResponder: BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587, type=HAS_DOWNSTREAM_IN_PIPELINE terminating shimmering.trading.imc.intra 2014-09-02 15:19:11,507 INFO org.apache.hadoop.hdfs.server.datanode.DataNode: PacketResponder: BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587, type=HAS_DOWNSTREAM_IN_PIPELINE terminating silverpine.trading.imc.intra 2014-09-02 15:19:11,508 INFO org.apache.hadoop.hdfs.server.datanode.DataNode.clienttrace: src: /10.141.0.155:52119, dest: /10.141.0.174:50010, bytes: 536000000, op: HDFS_WRITE, cliID: DFSClient_NONMAPREDUCE_1486044046_1, offset: 0, srvID: 8bc724e7-0f5c-4347-be32-7b9812286c5c, blockid: BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587, duration: 30358704352 tirisfal.trading.imc.intra 2014-09-02 15:19:11,508 INFO org.apache.hadoop.hdfs.server.datanode.DataNode: PacketResponder: BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587, type=HAS_DOWNSTREAM_IN_PIPELINE terminating sorrows.trading.imc.intra 2014-09-02 15:19:11,508 INFO org.apache.hadoop.hdfs.server.datanode.DataNode: PacketResponder: BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587, type=HAS_DOWNSTREAM_IN_PIPELINE terminating tirisfal.trading.imc.intra 2014-09-02 15:19:11,509 INFO org.apache.hadoop.hdfs.server.datanode.DataNode.clienttrace: src: /10.141.0.156:43356, dest: /10.141.0.170:50010, bytes: 536000000, op: HDFS_WRITE, cliID: DFSClient_NONMAPREDUCE_1486044046_1, offset: 0, srvID: 6f0e01d4-81a0-4289-b21a-4ca03bc6a191, blockid: BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587, duration: 30359566398 searing.trading.imc.intra 2014-09-02 15:19:11,509 INFO org.apache.hadoop.hdfs.server.datanode.DataNode.clienttrace: src: /10.141.0.170:35201, dest: /10.141.0.155:50010, bytes: 536000000, op: HDFS_WRITE, cliID: DFSClient_NONMAPREDUCE_1486044046_1, offset: 0, srvID: 08b14350-3484-4a0e-8520-8167733c62bd, blockid: BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587, duration: 30359128561 duskwood.trading.imc.intra 2014-09-02 15:19:11,509 INFO org.apache.hadoop.hdfs.server.datanode.DataNode: PacketResponder: BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587, type=HAS_DOWNSTREAM_IN_PIPELINE terminating duskwood.trading.imc.intra 2014-09-02 15:19:11,509 INFO org.apache.hadoop.hdfs.server.datanode.DataNode: PacketResponder: BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587, type=HAS_DOWNSTREAM_IN_PIPELINE terminating searing.trading.imc.intra 2014-09-02 15:19:11,510 INFO org.apache.hadoop.hdfs.server.datanode.DataNode.clienttrace: src: /10.141.0.157:36874, dest: /10.141.0.156:50010, bytes: 536000000, op: HDFS_WRITE, cliID: DFSClient_NONMAPREDUCE_1486044046_1, offset: 0, srvID: ddcdea62-71e0-4279-a96c-aeea90580249, blockid: BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587, duration: 30359733153 elwynn.trading.imc.intra 2014-09-02 15:19:11,510 INFO org.apache.hadoop.hdfs.server.datanode.DataNode.clienttrace: src: /10.141.0.159:46759, dest: /10.141.0.168:50010, bytes: 536000000, op: HDFS_WRITE, cliID: DFSClient_NONMAPREDUCE_1486044046_1, offset: 0, srvID: c953d2ac-57ba-47a4-ad49-24161a18770b, blockid: BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587, duration: 30360356635 redridge.trading.imc.intra 2014-09-02 15:19:11,510 INFO org.apache.hadoop.hdfs.server.datanode.DataNode.clienttrace: src: /10.141.0.168:37742, dest: /10.141.0.157:50010, bytes: 536000000, op: HDFS_WRITE, cliID: DFSClient_NONMAPREDUCE_1486044046_1, offset: 0, srvID: ab89db9e-c99c-4a6e-9e87-cec30626cef9, blockid: BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587, duration: 30360119140 eversong.trading.imc.intra 2014-09-02 15:19:11,510 INFO org.apache.hadoop.hdfs.server.datanode.DataNode: PacketResponder: BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587, type=HAS_DOWNSTREAM_IN_PIPELINE terminating elwynn.trading.imc.intra 2014-09-02 15:19:11,510 INFO org.apache.hadoop.hdfs.server.datanode.DataNode: PacketResponder: BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587, type=HAS_DOWNSTREAM_IN_PIPELINE terminating eversong.trading.imc.intra 2014-09-02 15:19:11,510 INFO org.apache.hadoop.hdfs.server.datanode.DataNode: PacketResponder: BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587, type=HAS_DOWNSTREAM_IN_PIPELINE terminating redridge.trading.imc.intra 2014-09-02 15:19:11,511 INFO org.apache.hadoop.hdfs.server.datanode.DataNode.clienttrace: src: /10.141.0.87:55341, dest: /10.141.0.159:50010, bytes: 536000000, op: HDFS_WRITE, cliID: DFSClient_NONMAPREDUCE_1486044046_1, offset: 0, srvID: 9ba59389-bce8-4cf0-815c-b96a20699d13, blockid: BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587, duration: 30328724294 hinterlands.trading.imc.intra 2014-09-02 15:19:11,511 INFO org.apache.hadoop.hdfs.server.datanode.DataNode: PacketResponder: BP-403847522-10.141.0.153-1407450167649:blk_1080927884_7187587, type=HAS_DOWNSTREAM_IN_PIPELINE terminating hinterlands.trading.imc.intra From: Stanley Shi <s...@pivotal.io<mailto:s...@pivotal.io>> Reply-To: "user@hadoop.apache.org<mailto:user@hadoop.apache.org>" <user@hadoop.apache.org<mailto:user@hadoop.apache.org>> Date: Tuesday, September 2, 2014 at 4:30 AM To: "user@hadoop.apache.org<mailto:user@hadoop.apache.org>" <user@hadoop.apache.org<mailto:user@hadoop.apache.org>> Cc: Julien Lehuen <julien.leh...@imc.nl<mailto:julien.leh...@imc.nl>>, Tyler McDougall <tyler.mcdoug...@imc.nl<mailto:tyler.mcdoug...@imc.nl>> Subject: Re: Replication factor affecting write performance What's the network setup and topology? Also, the size of the cluster? On Mon, Sep 1, 2014 at 4:10 PM, Laurens Bronwasser <laurens.bronwas...@imc.nl<mailto:laurens.bronwas...@imc.nl>> wrote: And now with the right label on the Y-axis. [cid:7D64F387-C6F1-4B37-9894-0166EC949EF9] From: Microsoft Office User <laurens.bronwas...@imc.nl<mailto:laurens.bronwas...@imc.nl>> Date: Monday, September 1, 2014 at 9:56 AM To: "user@hadoop.apache.org<mailto:user@hadoop.apache.org>" <user@hadoop.apache.org<mailto:user@hadoop.apache.org>> Cc: Julien Lehuen <julien.leh...@imc.nl<mailto:julien.leh...@imc.nl>>, Tyler McDougall <tyler.mcdoug...@imc.nl<mailto:tyler.mcdoug...@imc.nl>> Subject: Replication factor affecting write performance Hi, We have a setup with two clusters. On cluster shows a very strong degradation when we increase the replication factor. Another cluster shows hardly any degradation with increased replication factor. Any idea how to find out the bottleneck in the slower cluster? ________________________________ The information in this e-mail is intended only for the person or entity to which it is addressed. It may contain confidential and /or privileged material. If someone other than the intended recipient should receive this e-mail, he / she shall not be entitled to read, disseminate, disclose or duplicate it. If you receive this e-mail unintentionally, please inform us immediately by "reply" and then delete it from your system. Although this information has been compiled with great care, neither IMC Financial Markets & Asset Management nor any of its related entities shall accept any responsibility for any errors, omissions or other inaccuracies in this information or for the consequences thereof, nor shall it be bound in any way by the contents of this e-mail or its attachments. In the event of incomplete or incorrect transmission, please return the e-mail to the sender and permanently delete this message and any attachments. Messages and attachments are scanned for all known viruses. Always scan attachments before opening them. -- Regards, Stanley Shi, [http://www.gopivotal.com/files/media/logos/pivotal-logo-email-signature.png] ________________________________ The information in this e-mail is intended only for the person or entity to which it is addressed. It may contain confidential and /or privileged material. If someone other than the intended recipient should receive this e-mail, he / she shall not be entitled to read, disseminate, disclose or duplicate it. If you receive this e-mail unintentionally, please inform us immediately by "reply" and then delete it from your system. Although this information has been compiled with great care, neither IMC Financial Markets & Asset Management nor any of its related entities shall accept any responsibility for any errors, omissions or other inaccuracies in this information or for the consequences thereof, nor shall it be bound in any way by the contents of this e-mail or its attachments. In the event of incomplete or incorrect transmission, please return the e-mail to the sender and permanently delete this message and any attachments. Messages and attachments are scanned for all known viruses. Always scan attachments before opening them.