[Buildbot-devel] large variability in slave compile times -- trying to optimize

[Philippe McLean]

I think we may be swapping on the VM host ... I will update when I have more
information.

FTR, assigning multiple cores to a vm does not substantially improve build
times under devenv.com (even when multi processing is enabled).
With a 2-core vm (under vmware fusion), I see 8:32 build time; with 1 core,
I see 9:35 build time.




On Fri, Jun 24, 2011 at 4:18 PM, Dustin J. Mitchell <dustin at v.igoro.us>wrote:

> On Fri, Jun 24, 2011 at 5:25 PM, Philippe McLean
> <philippe.mclean at gmail.com> wrote:
> > I'm wondering if there might be some network/IO issues with buildbot on
> the
> > server? (Does anyone else run masters on windows?)
>
> I don't think many do..
>
> It'd be interesting to know *where* the slaves are spending most of
> their time.  How much log output do the jobs produce?  There should be
> enough buffer space for a spike in load on the master to not even
> cause the ongoing build process to block.
>
> I don't have any particular suggestions for debugging, since I would
> probably dive right in with strace or dtruss, and I dont' know if such
> tools are available for Windows.
>
> I've seen a lot of trouble with VMs and wildly divergent build times -
> I don't have any idea why, though.
>
> Can you add a bug for this, and use the keyword 'performance', to
> track your investigations?
>
> Dustin
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://buildbot.net/pipermail/devel/attachments/20110624/73d52eea/attachment.html>