[ltt-dev] performance results I measured with the latest lttng patches

[Nikhil Rao]

On Fri, Sep 26, 2008 at 9:43 AM, Martin Bligh <mbligh at google.com> wrote:

> > Yup, what I am currently doing (by hand, it's rather long), is to run
> > tbench between LTTng patches application to see where the performance
> > regressions are.
> >
> > Some automated build/test infrastructure would be good, e.g. :
> >
> > quilt pop -a
> >
> > until patchset fully applied
> >  quilt push
> >  build kernel
> >  install kernel
> >  reboot
> >  run tbench, gather results
> >
> > This should probably be done with both config=Y and config=N for
> > tracepoint and lttng-specific options.
>
> Autotest can do this for you. Nikhil (cc'ed) even has patches to do
> bisection chop search (though you'd need to make the test throw
> an exception below a certain perf level).
>

Yes, I have a bisection chop search test that can apply a patch (or
alternatively perform an action like 'quilt push'), build, install, reboot
and run tbench. Designing a test to raise an exception if tbench perf is
below a certain level is pretty simple. The tough part is deciding what this
threshold should be ;-) I can set this up for you if you'd like.

-Thanks,
Nikhil
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.casi.polymtl.ca/pipermail/lttng-dev/attachments/20080926/ec6ba2fb/attachment-0002.htm>