Performance testing framework

Jeremy Allison jra at samba.org
Tue Sep 27 16:43:29 UTC 2016


On Tue, Sep 27, 2016 at 05:48:26PM +1300, Douglas Bagnall wrote:
> A while ago I wrote:
> 
> > https://www.samba.org/~dbagnall/perf-tests/
> >
> > [...]
> >
> > The next trick I am working on is to run the tests multiple times and
> > use the fastest. Then I'll look at regular automatic testing of master.
> 
> I did those things. The picture at the URL below tracks a years worth
> of performance, looking at the autobuild-tagged commit closest to each
> Saturday (in some timezone). Each test run is completed 3 times, and
> the fastest result is used. I don't know why there are only two
> results from January.
> 
> https://www.samba.org/~dbagnall/perf-tests/master-52-weeks.html
> 
> If you've been following along this may be rather uninteresting. It
> again shows we suddenly got faster in a number of tests -- these tests
> were designed to show that we got faster there. Nevertheless the
> overall picture is good. Nowhere in this limited subset of AD
> functionality have we got slower in master over the last year. Some
> areas, like search, look like they have been getting quietly faster
> over several weeks.
> 
> My next plans for this, if I get time, are to set up a cron job that
> will keep this graph up to date, and switch to using some kind of
> modern javascript chart that is easier to unpick. Anyone who wants me
> to test the performance of other parts of Samba should send patches.
> The tests should look just like selftest tests that aren't going to
> fail and that take a measurable number of seconds to run.

This is great work Douglas - extremely interesting statistics !

Thanks a *LOT* for doing this !

Jeremy.



More information about the samba-technical mailing list