[
https://issues.apache.org/jira/browse/GEODE-6233?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16727083#comment-16727083
]
ASF subversion and git services commented on GEODE-6233:
--------------------------------------------------------
Commit d838b34034d20b08ab3730bd1e2f9288ea530170 in geode-benchmarks's branch
refs/heads/GEODE-6233 from Helena Bales
[ https://gitbox.apache.org/repos/asf?p=geode-benchmarks.git;h=d838b34 ]
GEODE-6233: Use a new hash-map for each PutAll batch in PrePopulateRegion
We see an occasional delay/hang when simply clearing and reusing the same
hashmap for multiple putAll requests. This change creates a new hashmap every
time it starts a new batch.
This also increases the keyRange for the benchmarks from 10k to 1M.
Co-Authored-By: Helena Bales <[email protected]>
> Prepopulate phase of benchmarks is taking too long
> --------------------------------------------------
>
> Key: GEODE-6233
> URL: https://issues.apache.org/jira/browse/GEODE-6233
> Project: Geode
> Issue Type: Bug
> Components: benchmarks
> Reporter: Brian Rowe
> Priority: Major
> Attachments: number of entries in 1M entry test.png
>
>
> We are seeing very strange behavior with the prepopulate phase of the
> benchmarks. With 10K keys, the prepopulate phase occasionally takes several
> minutes.
> Looking at the stats, we see that almost all of the data is loaded
> immediately, but there is a long tail of just a few new entries being created
> per second. This appears to be the result a single or very small number of
> PutAll operations having a hard time completing their last few values.
> When we bump up the number of keys to 1M, we see similar behavior, but the
> initial burst is an even lower fraction of the data, and we're still only
> around 500K entries created after an hour of prepopulation.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
