[
https://issues.apache.org/jira/browse/SUREFIRE-1516?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16452769#comment-16452769
]
Fork Script commented on SUREFIRE-1516:
---------------------------------------
Hi Tibor,
Thanks for the response. We did only some basic profiling, and we
found that non-negligible time goes into deserializing options from
file in ForkBooter, as well as inter-process communication (e.g., the
way threads wait for outputs from forked processes).
Let us know what would be a good next step. We obtained a lot more
results (both for forkCount=1, as well as forkCount=2 and forkCount=4,
and we find consistent speedup if we simply spawn processes). Note
that we do not claim that all the build configurations would simply
work with the suggested simple spawning, but for running things on CI
for basic configurations (when the only output that may matter is
pass/fail) seems to work nice.
Thanks
> Should surefire specialize test runner when test isolation (i.e., fork) is
> needed?
> ----------------------------------------------------------------------------------
>
> Key: SUREFIRE-1516
> URL: https://issues.apache.org/jira/browse/SUREFIRE-1516
> Project: Maven Surefire
> Issue Type: New Feature
> Components: Maven Surefire Plugin
> Affects Versions: 2.21.0
> Reporter: Fork Script
> Priority: Major
>
> We observed that {code:java} mvn test -DforkCount=1 -DreuseForks=false {code}
> has high overhead for several projects we recently used.
> We developed a specific runner for this option, which reduces test execution
> time by 49% on average. (Please see detailed results in the table below.)
> [Here|https://github.com/forkscript/maven-surefire/commit/38f102f0165f081dd28c6a3023e0b052662857f0]
> is a link to our commit (see ForkStarter.java:388). Our approach is hackish,
> but we believe it can be engineered better if needed. Is this of interest to
> surefire? We would be happy to discuss this further and provide more results.
> Detailed results are below. Machine we used: Intel i7-6700U CPU @ 3.40GHz
> with 16GB of RAM, running Ubuntu 17.10. We can share our script used to
> collect the numbers.
> || project || maven current (sec) || maven with our hack (sec) || time
> reduced by ||
> | activiti | 1300.54 | 1069.20 | 17% |
> | bukkit | 18.98 | 4.80 | 74% |
> | closurecompiler | 399.42 | 271.68 | 31% |
> | commonscodec | 46.98 | 13.85 | 70% |
> | commonsio | 107.35 | 67.41 | 37% |
> | commonslang | 89.70 | 35.39 | 60% |
> | commonsmath | 391.20 | 110.79 | 71% |
> | commonsnet | 74.50 | 59.46 | 20% |
> | commonsvalidator | 37.30 | 13.24 | 64% |
> | configuration | 126.73 | 62.37 | 50% |
> | crypto | 23.75 | 7.14 | 69% |
> | gedcom4j | 195.43 | 64.10 | 67% |
> | graphhopper | 90.80 | 41.25 | 54% |
> | jaxx | 36.00 | 23.33 | 35% |
> | la4j | 15.23 | 7.60 | 50% |
> | openfire | 28.52 | 18.51 | 35% |
> | retrofit | 53.22 | 35.36 | 33% |
> | scribejava | 13.35 | 4.32 | 67% |
> | vectorz | 37.90 | 11.19 | 70% |
> | ztexec | 23.94 | 17.75 | 25% |
> | Avg. | 155.54 | 96.93 | 49% |
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
