[jboss-jira] [JBoss JIRA] (WFLY-5822) Clustering performance regression in ejbremote-dist-sync scenario

[Richard Achmatowicz (JIRA)]

    [ https://issues.jboss.org/browse/WFLY-5822?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13150767#comment-13150767 ] 

Richard Achmatowicz commented on WFLY-5822:
-------------------------------------------

Also ran the ejbservlet stress test using repl-sync and based on the first comment in this JIRA issue, was expecting the performance of EAP7 to be as good or better than EAP6. However, the throughput of EAP6 is markedly better than that of EAP7 - 7.5K requests at 1200 clients for EAP6 versus 5.2K requests for 800 clients after which the curve flat lines. 

The early flat lining looks suspicious - maybe an upper limit is being hit elsewhere?
* looking at the network usage in the cluster, EAP6 usage goes well beyond 1,800,000 bytes/sec whereas EAP7 hits a ceiling of 1,350,000 bytes/sec. Same story for the driver nodes. So the network doesn't seem to be being used fully
* looking at CPU usage, the cluster node CPU usage for both EAP6 and EAP 7 are around the 55% - 65% mark; similarly, the driver nodes are around the 225-27% mark
* cluster node memory stats are not immediately available on the graph - it would be interesting to see what is going on here, as CG could be at work to slow things down

One thing I find odd here: the EAP 6 tests are taking about half the number of iterations as the EAP 7 tests. For example, in the ejbservlet-repl-sync tests, we have 8 iterations for EAP6 and 15 iterations for EAP7. Wonder why? Is it because the EAP7 performs slower and so does not hit the stop threshold sooner? 

> Clustering performance regression in ejbremote-dist-sync scenario 
> ------------------------------------------------------------------
>
>                 Key: WFLY-5822
>                 URL: https://issues.jboss.org/browse/WFLY-5822
>             Project: WildFly
>          Issue Type: Bug
>          Components: Clustering, EJB
>    Affects Versions: 10.0.0.CR5
>            Reporter: Michal Vinkler
>            Assignee: Richard Achmatowicz
>            Priority: Critical
>
> Compared to EAP 6, all SYNC scenarios have the same/better performance except of this one, wonder why?
> Compare these results:
> stress-ejbremote-dist-sync
> 7.0.0.ER2: [throughput|http://jenkins.mw.lab.eng.bos.redhat.com/hudson/job/eap-7x-stress-ejbremote-dist-sync/4/artifact/report/graph-throughput.png]
> 6.4.0.GA: [throughput|http://jenkins.mw.lab.eng.bos.redhat.com/hudson/job/eap-6x-stress-ejbremote-dist-sync_noperf21/1/artifact/report/graph-throughput.png]
> ---------------------------------------
> Just for comparison: ejbremote REPL_SYNC scenario *performs well* on the other hand:
> stress-ejbremote-repl-sync
> 7.0.0.ER2: [throughput|http://jenkins.mw.lab.eng.bos.redhat.com/hudson/job/eap-7x-stress-ejbremote-repl-sync/3/artifact/report/graph-throughput.png]
> 6.4.0.GA: [throughput|http://jenkins.mw.lab.eng.bos.redhat.com/hudson/job/eap-6x-stress-ejbremote-repl-sync_noperf21/2/artifact/report/graph-throughput.png]



--
This message was sent by Atlassian JIRA
(v6.4.11#64026)