[
https://issues.apache.org/jira/browse/HADOOP-2533?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
stack updated HADOOP-2533:
--------------------------
Attachment: dirty.patch
Attached dirty patch -- it needs a bunch of work to make it committable --
shrinks MapWritable CPU usage over a PerformanceEvaluation run from 22.9% of
all CPU used down to 0.8%. It uses a dumber MapWritable -- one that has static
maps of classes to codes -- and one that is more friendly to hbase; it has
been primed with hbase type (they have to be dynamically 'added' to each new
instantiation of the hadoop.io version of MapWritable).
> [hbase] Performance: Scanning, just creating MapWritable in next consumes
> >20% CPU
> ----------------------------------------------------------------------------------
>
> Key: HADOOP-2533
> URL: https://issues.apache.org/jira/browse/HADOOP-2533
> Project: Hadoop
> Issue Type: Improvement
> Components: contrib/hbase
> Reporter: stack
> Priority: Minor
> Attachments: dirty.patch
>
>
> Every call to HScanner.next creates an instance of MapWritable. MapWritables
> are expensive. Watching a scan run in the profiler, the setup of the
> MapWritable -- filling out the idToClassMap and classToIdMap -- consumes 20%
> of all CPU.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
