[
https://issues.apache.org/jira/browse/HADOOP-2768?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12564974#action_12564974
]
Raghu Angadi commented on HADOOP-2768:
--------------------------------------
+1.
Yes, this will increase buffer copies. But we intend get rid of copies without
changing effective write size for sockets for 0.17 (HADOOP-2154). I went
through the same experience with HADOOP-1134, initially didn't have
BufferedOutputStream and added it after the benchmarks.
> DFSIO write performance benchmark shows a regression
> ----------------------------------------------------
>
> Key: HADOOP-2768
> URL: https://issues.apache.org/jira/browse/HADOOP-2768
> Project: Hadoop Core
> Issue Type: Bug
> Components: dfs
> Affects Versions: 0.16.0
> Reporter: Mukund Madhugiri
> Assignee: dhruba borthakur
> Priority: Blocker
> Fix For: 0.16.0
>
> Attachments: perf.patch, perf.patch
>
>
> We are seeing a performance regression in DFSIO write benchmark, possibly due
> to Hadoop-1707.
> Here is the data:
> - Aggregate cluster throughput went from 0.30G to 0.14G with 39 concurrent
> clients, on 20 nodes
> - Aggregate cluster throughput went from 1.15G to 0.85G with 39 concurrent
> clients, on 100 nodes
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
