Re: [Monotone-devel] cvs_import benchmark

[Nathaniel Smith]

On Thu, Oct 12, 2006 at 08:31:40AM +0200, Lapo Luchini wrote:
> % du -hs cvsroot.ypsilon
> 1.5M    cvsroot.ypsilon
> 
> % python2.4 benchmark.py -m mtn26=mtn-0.26.exe -m mtn30=mtn-0.30.exe -m
> mtn061011=mtn-061011.exe -b 'cvsimp=CVSImport(EmptyRepo(), "ypsilon",
> "/home/lapo/monotone-trunk/benchmark/cvsroot.ypsilon")' --cache
> /home/lapo/monotone-trunk/benchmark/cache scratch results3
> 
> % for n in results3/*/stats.csv; do echo $n; cat $n; done
> results3/cvsimp-mtn26-time/stats.csv
> cvs_import-system-time,0.51,0.53,0.59
> cvs_import-user-time,104.54,105.25,103.37
> cvs_import-wall-time,130.21,127.43,127.83
> results3/cvsimp-mtn30-time/stats.csv
> cvs_import-system-time,0.46,0.48,0.49
> cvs_import-user-time,97.07,98.1,97.09
> cvs_import-wall-time,119.28,119.21,119.68
> results3/cvsimp-mtn061011-time/stats.csv
> cvs_import-system-time,0.79,0.54,0.81
> cvs_import-user-time,196.57,196.5,192.68
> cvs_import-wall-time,236.94,234.7,230.96
> 
> So... 6% improvement between 0.26 and 0.30, and a 100% worsening from
> 0.30 and yesterday's trunk! 0_0

Weird!

> I wonder what can have done that. Maybe 'heights' branch is landed and
> was not faster as expected?

No, it hasn't landed yet...

> I'll definitely take a closer look at it this evening, now gotta go to
> work ;-)

On unix, obviously, we could quickly peek into the problem by adding
some of the profiling switches, but I dunno what you can do on
windows.

-- Nathaniel

-- 
- Don't let your informants burn anything.
- Don't grow old.
- Be good grad students.
  -- advice of Murray B. Emeneau on the occasion of his 100th birthday