[
https://issues.apache.org/jira/browse/HDDS-3223?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
runzhiwang updated HDDS-3223:
-----------------------------
Description:
*What's the problem ?*
Read a 300M file, it cost about 25 seconds, i.e. 12M/s, which is too slow. Then
I capture the packet. You can see from the image, read a 300M file need 10 GET
requests, each GET request read about 30M.
The first GET request cost about 1 second, but the 10th GET request cost about
23 seconds.
!screenshot-1.png!
*What's the reason ?*
When do GET, the stack is:
[IOUtils::copyLarge|https://github.com/apache/hadoop-ozone/blob/master/hadoop-ozone/s3gateway/src/main/java/org/apache/hadoop/ozone/s3/endpoint/ObjectEndpoint.java#L262]
->
[IOUtils::skipFully|https://github.com/apache/commons-io/blob/master/src/main/java/org/apache/commons/io/IOUtils.java#L1190]
->
[IOUtils::skip|https://github.com/apache/commons-io/blob/master/src/main/java/org/apache/commons/io/IOUtils.java#L2064]
->
[InputStream::read|https://github.com/apache/commons-io/blob/master/src/main/java/org/apache/commons/io/IOUtils.java#L1957].
It means, the 10th GET request which should read 270M-300M, but to skip 0-270M,
it also
[InputStream::read|https://github.com/apache/commons-io/blob/master/src/main/java/org/apache/commons/io/IOUtils.java#L1957]
0-270M. So the GET request become slower and slower
You can also refer it [here|https://issues.apache.org/jira/browse/IO-355] why
IOUtils implement skip by read rather than real skip, e.g. seek.
was:
*What's the problem ?*
Read a 300M file, it cost about 25 seconds, i.e. 12M/s, which is too slow. Then
I capture the packet. You can see from the image, read a 300M file need 10 GET
requests, each GET request read about 30M.
The first GET request cost about 1 second, but the 10th GET request cost about
23 seconds.
!screenshot-1.png!
*What's the reason ?*
When do GET, the stack is:
[IOUtils::copyLarge|https://github.com/apache/hadoop-ozone/blob/master/hadoop-ozone/s3gateway/src/main/java/org/apache/hadoop/ozone/s3/endpoint/ObjectEndpoint.java#L262]
->
[IOUtils::skipFully|https://github.com/apache/commons-io/blob/master/src/main/java/org/apache/commons/io/IOUtils.java#L1190]
->
[IOUtils::skip|https://github.com/apache/commons-io/blob/master/src/main/java/org/apache/commons/io/IOUtils.java#L2064]
->
[InputStream::read|https://github.com/apache/commons-io/blob/master/src/main/java/org/apache/commons/io/IOUtils.java#L1957].
It means, the 10th GET request which should read 270M-300M, but to skip 0-270M,
it also
[InputStream::read|https://github.com/apache/commons-io/blob/master/src/main/java/org/apache/commons/io/IOUtils.java#L1957]
0-270M. So the GET request become slower and slower
> S3g become slower when read bigger object for error use of skip
> ---------------------------------------------------------------
>
> Key: HDDS-3223
> URL: https://issues.apache.org/jira/browse/HDDS-3223
> Project: Hadoop Distributed Data Store
> Issue Type: Improvement
> Reporter: runzhiwang
> Assignee: runzhiwang
> Priority: Critical
> Attachments: screenshot-1.png
>
>
> *What's the problem ?*
> Read a 300M file, it cost about 25 seconds, i.e. 12M/s, which is too slow.
> Then I capture the packet. You can see from the image, read a 300M file need
> 10 GET requests, each GET request read about 30M.
> The first GET request cost about 1 second, but the 10th GET request cost
> about 23 seconds.
> !screenshot-1.png!
> *What's the reason ?*
> When do GET, the stack is:
> [IOUtils::copyLarge|https://github.com/apache/hadoop-ozone/blob/master/hadoop-ozone/s3gateway/src/main/java/org/apache/hadoop/ozone/s3/endpoint/ObjectEndpoint.java#L262]
> ->
> [IOUtils::skipFully|https://github.com/apache/commons-io/blob/master/src/main/java/org/apache/commons/io/IOUtils.java#L1190]
> ->
> [IOUtils::skip|https://github.com/apache/commons-io/blob/master/src/main/java/org/apache/commons/io/IOUtils.java#L2064]
> ->
> [InputStream::read|https://github.com/apache/commons-io/blob/master/src/main/java/org/apache/commons/io/IOUtils.java#L1957].
> It means, the 10th GET request which should read 270M-300M, but to skip
> 0-270M, it also
> [InputStream::read|https://github.com/apache/commons-io/blob/master/src/main/java/org/apache/commons/io/IOUtils.java#L1957]
> 0-270M. So the GET request become slower and slower
> You can also refer it [here|https://issues.apache.org/jira/browse/IO-355] why
> IOUtils implement skip by read rather than real skip, e.g. seek.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: ozone-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: ozone-issues-h...@hadoop.apache.org
