[
https://issues.apache.org/jira/browse/CASSANDRA-1369?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Brandon Williams updated CASSANDRA-1369:
----------------------------------------
Attachment: 1369.txt
Patch to use the threadlocal approach.
> FBUtilities.hash can result in thread contention on call to
> MessageDigest.getInstance()
> ---------------------------------------------------------------------------------------
>
> Key: CASSANDRA-1369
> URL: https://issues.apache.org/jira/browse/CASSANDRA-1369
> Project: Cassandra
> Issue Type: Improvement
> Components: Core
> Reporter: Jason Fager
> Assignee: Brandon Williams
> Priority: Minor
> Fix For: 0.7.1
>
> Attachments: 1369.txt
>
>
> FBUtilities.hash() calls MessageDigest.getInstance() on every invocation,
> which in turns calls the synchronized method Provider.getService().
> FBUtilities.md5hash() is frequently invoked from RandomPartitioner, and minor
> thread contention in this codepath can be observed when running
> contrib/py_stress against an out-of-box Cassandra installation.
> One possible solution is to preallocate md5 MessageDigest instances and store
> them as threadlocals.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
