hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Suresh Srinivas (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDFS-347) DFS read performance suboptimal when client co-located on nodes with data
Date Fri, 05 Apr 2013 04:12:16 GMT

    [ https://issues.apache.org/jira/browse/HDFS-347?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13623343#comment-13623343
] 

Suresh Srinivas commented on HDFS-347:
--------------------------------------

I have refrained from commenting. What I have seen is lack of understanding, not being receptive
to suggestions, and complete disregard for the comments posted by a long term committer.

bq. The only purpose of the patch with 'Jenkins' in the name is to get a Jenkins run. It is
not for review. It was generated with 'git diff master' run from the HDFS-347 branch. You
could do a similar thing with subverison by checking out two copies and diffing them. I recommend
looking at the commits in subversion or git.
What is the basis of this statement? I had several times mentioned in the voting thread that,
I choose to review merge patch with several reasons why it works for me. Still you make the
statement that "The only purpose of the patch with 'Jenkins' in the name is to get a Jenkins
run". 

Repeated requests for generating clean patch has been ignored. 

Yes, you need to generate the right patch. Not just a patch that is close to the right patch.
If the patch is not the right one, Jenkins +1 has no meaning. Many comments have tried to
indicate this, several times [here|https://issues.apache.org/jira/browse/HDFS-347?focusedCommentId=13572860&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13572860],
[here|https://issues.apache.org/jira/browse/HDFS-347?focusedCommentId=13620632&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13620632]
and [here|https://issues.apache.org/jira/browse/HDFS-347?focusedCommentId=13621601&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13621601].
If this is not clear, you could ask others to guide you on how to do it (though the helpful
tips on how to do it are ignored). 

bq. can you please add these style comments to the style cleanup JIRA HDFS-4661? Loading this
JIRA makes my web browser slow to a crawl.
I can load this on my mobile device over wireless connection. No problem. If the comments
are related to the clean merge patch, it should be made in this jira, isn't it?

bq. The difference between my patch and yours is this...
After describing how to do it, you chose to ignore. After it is done and works, you seem to
be saying it does not count because the different is not significant? It does not matter whether
it is a few lines of difference or one line. Right merge patch needs to be generated.

In the interest of making progress on this issue, [~tlipcon] can you please help merge the
patch with cleanly and correctly generated merge patch? If you are busy, I will work with
[~szetszwo] on getting patch merged to trunk. [~szetszwo] can you please indicate if a +1
from Jenkins on a clean merge patch is sufficient to merge this change to trunk or you would
like to see any more changes?


                
> DFS read performance suboptimal when client co-located on nodes with data
> -------------------------------------------------------------------------
>
>                 Key: HDFS-347
>                 URL: https://issues.apache.org/jira/browse/HDFS-347
>             Project: Hadoop HDFS
>          Issue Type: Improvement
>          Components: datanode, hdfs-client, performance
>            Reporter: George Porter
>            Assignee: Colin Patrick McCabe
>         Attachments: 2013.01.28.design.pdf, 2013.01.31.consolidated2.patch, 2013.01.31.consolidated.patch,
2013.02.15.consolidated4.patch, 2013-04-01-jenkins.patch, all.tsv, a.patch, BlockReaderLocal1.txt,
full.patch, HADOOP-4801.1.patch, HADOOP-4801.2.patch, HADOOP-4801.3.patch, HDFS-347-016_cleaned.patch,
HDFS-347.016.patch, HDFS-347.017.clean.patch, HDFS-347.017.patch, HDFS-347.018.clean.patch,
HDFS-347.018.patch2, HDFS-347.019.patch, HDFS-347.020.patch, HDFS-347.021.patch, HDFS-347.022.patch,
HDFS-347.024.patch, HDFS-347.025.patch, HDFS-347.026.patch, HDFS-347.027.patch, HDFS-347.029.patch,
HDFS-347.030.patch, HDFS-347.033.patch, HDFS-347.035.patch, HDFS-347-branch-20-append.txt,
hdfs-347-merge.txt, hdfs-347-merge.txt, hdfs-347-merge.txt, hdfs-347.png, hdfs-347.txt, local-reads-doc
>
>
> One of the major strategies Hadoop uses to get scalable data processing is to move the
code to the data.  However, putting the DFS client on the same physical node as the data blocks
it acts on doesn't improve read performance as much as expected.
> After looking at Hadoop and O/S traces (via HADOOP-4049), I think the problem is due
to the HDFS streaming protocol causing many more read I/O operations (iops) than necessary.
 Consider the case of a DFSClient fetching a 64 MB disk block from the DataNode process (running
in a separate JVM) running on the same machine.  The DataNode will satisfy the single disk
block request by sending data back to the HDFS client in 64-KB chunks.  In BlockSender.java,
this is done in the sendChunk() method, relying on Java's transferTo() method.  Depending
on the host O/S and JVM implementation, transferTo() is implemented as either a sendfilev()
syscall or a pair of mmap() and write().  In either case, each chunk is read from the disk
by issuing a separate I/O operation for each chunk.  The result is that the single request
for a 64-MB block ends up hitting the disk as over a thousand smaller requests for 64-KB each.
> Since the DFSClient runs in a different JVM and process than the DataNode, shuttling
data from the disk to the DFSClient also results in context switches each time network packets
get sent (in this case, the 64-kb chunk turns into a large number of 1500 byte packet send
operations).  Thus we see a large number of context switches for each block send operation.
> I'd like to get some feedback on the best way to address this, but I think providing
a mechanism for a DFSClient to directly open data blocks that happen to be on the same machine.
 It could do this by examining the set of LocatedBlocks returned by the NameNode, marking
those that should be resident on the local host.  Since the DataNode and DFSClient (probably)
share the same hadoop configuration, the DFSClient should be able to find the files holding
the block data, and it could directly open them and send data back to the client.  This would
avoid the context switches imposed by the network layer, and would allow for much larger read
buffers than 64KB, which should reduce the number of iops imposed by each read block operation.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Mime
View raw message