[
https://jira.nuxeo.com/browse/NXBT-356?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=109438#comment-109438
]
Mathieu Guillaume commented on NXBT-356:
----------------------------------------
For the second run of tests, I tried to address the case of functional tests.
The one used was nuxeo-distribution-dm/ftest/selenium (offline) as we don't
have any significant WebDriver tests yet.
JDK used was Sun's Java 6, with MAVEN_OPTS="-Xmx1024m -XX:MaxPermSize=256m".
For Nuxeo, the default JAVA_OPTS are "-Xms512m -Xmx1024m -XX:MaxPermSize=512m"
(plus other non-memory-related ones).
For selenium, no memory options are passed by default.
* Server 1:
Due to a too-recent version of Firefox, the tests didn't pass, the results are
not significant (too many waits for timeout, results were around 33 minutes).
* Server 2:
Run times: 24m18, 26m29, 24m40
Alternative test - double maven memory: 25m02
Alternative test - double nuxeo memory: 25m09
Alternative test - 1GB selenium memory: 25m13
* Server 3:
Run times: 28m05, 28m11, 27m55
Alternative test - double maven memory: 27m48
Alternative test - double nuxeo memory: 27m40
* Server 4:
Run times: 25m11, 25m27, 24m57
Alternative test - double maven memory: 25m06
Alternative test - double nuxeo memory: 26m08
Conclusion:
- The test times are pretty much the same whatever the options used, the
defaults are high enough.
- The virtual machine (Server 3) is slightly slower, but not by much.
The limiting factor in this test seems to be Firefox, which was using the most
CPU during the tests.
The similar times on servers with a different number of cores can probably be
explained by the fact the Firefox is not optimized for multi-core yet: it's
multi-threaded, but most of the work happens in a single thread. Cf. [this blog
post|http://blog.mozilla.com/products/2011/07/15/goals-for-multi-process-firefox/]
for details.
> Define a CI slave for performance
> ---------------------------------
>
> Key: NXBT-356
> URL: https://jira.nuxeo.com/browse/NXBT-356
> Project: Nuxeo ECM Build/Test Environment
> Issue Type: Task
> Components: Continuous integration
> Reporter: Julien Carsique
> Assignee: Mathieu Guillaume
>
> Slave build performances
> Currently the build on the Jenkins slaves is about the same time than on a
> dev laptop.
> => we can make better !
> It's true that a big part of the Jenkins slave are poor hardware (the dwarf).
> Before taking any decision, we should probably do some profiling to see
> exactly what are the real impacting factors :
> test build on a bigger CPU (like Octo)
> use RAMFS to speedup I/O
> tweak compile options ?
> test on big RackSpace VMs
> => based on that we will be able to define what kind of Build slaves we need
> (then we will have to see what we can host internaly and what must be
> external)
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
_______________________________________________
ECM-tickets mailing list
ECM-tickets@lists.nuxeo.com
http://lists.nuxeo.com/mailman/listinfo/ecm-tickets
