[
https://issues.apache.org/jira/browse/AMQ-3557?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13416048#comment-13416048
]
Torsten Mielke commented on AMQ-3557:
-------------------------------------
There was recently an improvement made to the JDBC persistence adapter to
remove an unneeded synchronization point. See the last comment of AMQ-2868.
This fix improves performance of the JDBC persistence adapter in case of using
multiple consumers.
> Performance of consumption with JDBC persistance and Microsoft SQL Server
> -------------------------------------------------------------------------
>
> Key: AMQ-3557
> URL: https://issues.apache.org/jira/browse/AMQ-3557
> Project: ActiveMQ
> Issue Type: Bug
> Components: Message Store, Performance Test
> Affects Versions: 5.4.3, 5.5.0, 5.5.1
> Environment: Microsoft SQL Server 2005, Debian Linux,
> Reporter: Nicholas Rahn
> Labels: jdbc, performance, sqlserver
> Attachments: activemq.xml
>
>
> We are trying to upgrade our ActiveMQ installation and have run into some
> performance issues. I'll attached our activemq.xml file to this bug.
> I've setup a fresh SQLServer database for our upgrade tests and using the
> example Ant tools in the distribution, I've populated a persistent queue with
> 1,000,000 messages. I then consume those messages using the example Ant
> consumption script. The producing side works fine. However the performance of
> the consumption side is extremely poor. To consume just 10,000 of those
> messages takes over 5 minutes.
> The consumer will pause for 4-5 seconds every 200 messages. This is easily
> visible in the output of the Ant script. We have also traced the DB to see
> what is happening there and have found that the findNextMessagesStatement
> takes 4-5 seconds every time it is executed. The statement's ID parameter is
> increased by 200 every time it is executed. We also noticed the use of the
> "SET ROWCOUNT 10000" statement setting the maximum number of rows returned
> from a query at 10000. We also traced previous versions of ActiveMQ and found
> that SET ROWCOUNT was used much more often, with much smaller values (often
> 10, 20 or 30).
> We have also tested the same setup with version 5.4.0 and did not have the
> same issues. Consumption speeds with 5.4.0 were normal, with no pauses.
> Version 5.4.3 did have the problem, however. So there seems to be a
> regression somewhere between 5.4.0 and 5.4.3 (also affects 5.5.0 and later).
> Please let me know if you need more information, including the database
> traces.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
