hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "stack (JIRA)" <j...@apache.org>
Subject [jira] Commented: (HBASE-986) Cache sockets in DFSClient
Date Fri, 07 Nov 2008 07:52:44 GMT

    [ https://issues.apache.org/jira/browse/HBASE-986?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12645710#action_12645710
] 

stack commented on HBASE-986:
-----------------------------

The above 4X is a misreporting; I had a big cache configured when running tests.  Will be
back with proper numbers.

> Cache sockets in DFSClient
> --------------------------
>
>                 Key: HBASE-986
>                 URL: https://issues.apache.org/jira/browse/HBASE-986
>             Project: Hadoop HBase
>          Issue Type: Improvement
>            Reporter: stack
>            Assignee: stack
>            Priority: Blocker
>             Fix For: 0.19.0
>
>         Attachments: socketCache.patch
>
>
> Talking with Dhruba here at Apachecon, he suggested that setup of sockets is big price
payed during random-read.  Stefan hacked into DFSClient an ugly patch that would allow us
quickly test random-reading on top of a cache of cached sockets.  I just tested it.  I see
4X improvement.
> Making this a blocker on 0.19.0.  Looks like we can hack in a real socket pool using
a custom SocketFactory under DFSClient.  It'll be messy but the promised improvement is big.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Mime
View raw message