[
https://issues.apache.org/jira/browse/AMQ-3454?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13410271#comment-13410271
]
William commented on AMQ-3454:
------------------------------
I have integrated those changes (first mutex and then the remove of the copy of
the map) in a 5.5.1 and it works a lot better under load. This may not be the
best pattern by it makes ActiveMQ usable and more performant.
I have tested the 2 fix successively and my findings are :
* chaning mutex by read/write lock remove contentions between thread and
improve performance a lot (no more red on theads in jvisualwm)
* removing the copy of the map improve memory usage and performance
Tested with 2000 topics on a single broker with one message every 5 seconds.
Without those fixes it wasn't performant.
> Contention on a mutex during a stress when using SimpleAuthenticationPlugin
> ---------------------------------------------------------------------------
>
> Key: AMQ-3454
> URL: https://issues.apache.org/jira/browse/AMQ-3454
> Project: ActiveMQ
> Issue Type: Bug
> Components: Broker
> Affects Versions: 5.5.0
> Reporter: William
> Assignee: Gary Tully
> Priority: Critical
> Fix For: 5.6.0
>
> Attachments: RWLockDestinations.txt, amq-config.zip, amq-dump
>
>
> I am testing ActiveMQ 5.5 with my own stress test. I tried to implement a
> stress test that use jms ressource in the same fashion that my application
> would do.
> For that, I used jms template (from Spring) and a pooled connection factory
> (as recommended). I run a fixed number of thread that plublish on fixed
> number of topics. Each thread pick up a topic and enter a loop that will send
> a message on the choosen topic.
> The issue is reproductible with SimpleAuthenticationPlugin active.
> Configuration of the test :
> * 100 topics
> * more than one thread per topic (actually 1500 producer threads)
> Common connection properties :
> * alwaysSessionAsync=false
> * dispatchAsync=false
> * optimizeAcknowledge=true
> *
> socketBufferSize=131072&trace=true&wireFormat.cacheSize=2048&wireFormat.tcpNoDelayEnabled=true&wireFormat.tightEncodingEnabled=true&keepAlive=true&soTimeout=10000&connectionTimeout=10000\
> Producers :
> * Message are not persistent
> * There is no transaction
> * Message expiration time is 30s
> * Unsing JMS template singleton with a PooledConnectionFactory
> * 3 JVM running
> * connection options alwaysSyncSend=true,copyMessageOnSend=false
> Consumers :
> * just listening to each topic with a SimpleMessageListenerContainer and
> count messages received
> * 3 JVM running
> In attachment, you will find activemq config file and thread dumps. I can
> attach the producer/consumers code if you need it but I think you have your
> own tests.
> This Jira is critical for me because security is mandatory for our use.
> If you need more information, please ask.
--
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
