[
https://issues.apache.org/activemq/browse/AMQ-1386?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Rob Davies resolved AMQ-1386.
-----------------------------
Resolution: Working as Designed
Fix Version/s: (was: 5.4.0)
5.3.0
sync to disk on a transaction boundary is going to be slow. You can configure
not to do this - depends on the store used
> Very slow on Solaris/Linux when commiting transactions
> ------------------------------------------------------
>
> Key: AMQ-1386
> URL: https://issues.apache.org/activemq/browse/AMQ-1386
> Project: ActiveMQ
> Issue Type: Bug
> Affects Versions: 4.1.1, 5.2.0
> Environment: Solaris 10 on T2000, SUSE Enterprise linux 10 on DELL
> AMD64, Debian testing on AMD Turion laptop, SUSE8 on DELL amd
> Reporter: robert huitema
> Assignee: Rob Davies
> Fix For: 5.3.0
>
> Attachments: JmsTestHarnessAMQ.java
>
>
> All these systems show a very severe slowdown in TPS when each read is
> commited. this varies with the system, but is generally 100 times less than
> when I commit every 1000 reads. I have investigated and applied the fixes
> found so far, including the tcp_nodelay, net.ipv4.tcp_abc, and misc system
> fixes. The problem exhibits across localhost and network, and on Seebeyonds
> stcms IQManager.
> I have attached a test harness which should show the problem.
> Summary of speeds:
> localhost windows desktop PC: commit 1000: 3900 tps, commit each:1300 tps
> DELL AMD64: commit 1000: 8000tps, commit each: 30tps
> SunT2000: commit 1000:700tps, commit each; 15tps
> linux laptop: commit 1000: 2500 tps, commit each: 200 tps
> Hope you can help, we have been at this in depth for a while now :-(
> rgds
> rob
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
