hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "stack (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-9953) PerformanceEvaluation: Decouple data size from client concurrency
Date Mon, 28 Apr 2014 23:37:15 GMT

    [ https://issues.apache.org/jira/browse/HBASE-9953?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13983736#comment-13983736
] 

stack commented on HBASE-9953:
------------------------------

This seems to work nicely for me.  Funny how when the block cache is 4G, if you set the size
to get to 4G, you fall out of BC.  If you set it to 3G, you still do.  2.8G gets the BC 100%
used w/ no misses.  We've lots of overhead it seems (smile).  Let me remove my modulo malarky.

> PerformanceEvaluation: Decouple data size from client concurrency
> -----------------------------------------------------------------
>
>                 Key: HBASE-9953
>                 URL: https://issues.apache.org/jira/browse/HBASE-9953
>             Project: HBase
>          Issue Type: Test
>          Components: Performance
>            Reporter: Nick Dimiduk
>            Assignee: Nick Dimiduk
>            Priority: Minor
>             Fix For: 0.99.0
>
>         Attachments: HBASE-9953.00.patch, HBASE-9953.01.patch
>
>
> PerfEval tool provides a {{--rows=R}} for specifying the number of records to work with
and requires the user provide a value of N, used as the concurrency level. From what I can
tell, every concurrent process will interact with R rows. In order to perform an apples-to-apples
test, the user must re-calculate the value R for every new value of N.
> Instead, I propose accepting a {{--size=S}} for the amount of data to interact with and
let PerfEval divide that amongst the N clients on the user's behalf.



--
This message was sent by Atlassian JIRA
(v6.2#6252)

Mime
View raw message