hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jean-Daniel Cryans (JIRA)" <j...@apache.org>
Subject [jira] Updated: (HBASE-887) Fix a hotspot in scanners
Date Wed, 17 Sep 2008 19:27:44 GMT

     [ https://issues.apache.org/jira/browse/HBASE-887?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Jean-Daniel Cryans updated HBASE-887:
-------------------------------------

    Attachment: hbase-887-v1.patch

This patch fixes the hotspot. When running the PE with only 1 row prefetched (so the same
as usual), I got these numbers:

bq. Finished scan in 84713ms at offset 0 for 1048576 rows

With my patch in place and a configured 2000 rows of caching, I got:

bq. Finished scan in 34754ms at offset 0 for 1048576 rows

Passes the tests. Please try it and review. Maybe the default hbase.client.scanner.caching
should be 1 to be sure and a bigger number for MR jobs?

> Fix a hotspot in scanners
> -------------------------
>
>                 Key: HBASE-887
>                 URL: https://issues.apache.org/jira/browse/HBASE-887
>             Project: Hadoop HBase
>          Issue Type: Improvement
>          Components: client, regionserver
>            Reporter: Jean-Daniel Cryans
>            Assignee: Jean-Daniel Cryans
>             Fix For: 0.19.0
>
>         Attachments: hbase-887-v1.patch
>
>
> When scanning, we do a lot of RPCs and this has a huge performance hit. I propose that
we add a way to fetch more rows during next() and put them in cache. This should be configurable.

-- 
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