[
https://issues.apache.org/jira/browse/HADOOP-3272?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12590572#action_12590572
]
shv edited comment on HADOOP-3272 at 4/18/08 11:58 AM:
-----------------------------------------------------------------------
The patch is fairly simple, but might have consequences.
I am all for committing it but before we do lets run some benchmarks.
I propose to run TestDFSIO on at least a 100 nodes cluster creating say 400
files with 100 small blocks each simultaneously with the patch an without.
It would be very useful to post jmap -histo outputs for Block and BlockInfo
objects before and after the patch is applied.
As well as some estimates of what the average gain per block is going to be.
was (Author: shv):
The patch is fairly simple, but might have consequences.
I am all for committing it but before we do lets run some benchmarks.
I propose to run TestDFSIO on at least a 100 nodes cluster creating say 400
files with 100 small blocks each simultaneously with the patch an without.
It would be very useful to post jmap -histo outputs for Block and BlockInfo
objects before and after the patch is applied.
> Reduce redundant copy of Block object in BlocksMap.map hash map
> ---------------------------------------------------------------
>
> Key: HADOOP-3272
> URL: https://issues.apache.org/jira/browse/HADOOP-3272
> Project: Hadoop Core
> Issue Type: Bug
> Components: dfs
> Environment: All
> Reporter: lohit vijayarenu
> Assignee: lohit vijayarenu
> Fix For: 0.18.0
>
> Attachments: HADOOP-3272.patch
>
>
> Looks like we might have redundant copy of Block object as Key for
> BlocksMap.map hash map. We should restore this to using same object for both
> Key, Value to save space.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
