On 2013-01-31 01:05, Niels Thykier wrote:
> On 2013-01-30 14:12, Niels Thykier wrote:
>> [...]
>
Just adding the current numbers so I don't forget them (and don't
compare against the wrong/old results). :)
> With this second patch on top of the previous one, the t/tests suite
> goes from:
>
> real 9m22.148s
> user 58m8.566s
> sys 6m45.349s
>
> to:
>
> [...]
7m34.921s / 43m33.755s / 6m9.935s
(That was the worst of the 3 runs I witnessed. Its best run was at
6m15s / 34m11s / 4m54s - but that leap makes it hard to trust, so I went
with the worst.)
> The kfreebsd-utils packages (15 binaries, 5 udebs, 1 dsc) goes from 51s
> to 28 user time (at -j9). At -j1 I can now process them at 35s-36s
> instead of 51s wall time.
>
At 20-25s / 25-26s / 2s now for -j9. With -j1 I see 34-35s / 25s / 2s.
>>
>> [...]
>> To get a feeling of the improvement works on real data we get...
>>
>
> fields-wrong-section:
> * master branch: 14.815s 32.818s 4.628s
> * first patch: 13.259s 20.733s 3.312s
> * both patches: 10.876s 10.257s 3.068s
>
10.3s/6.8s/2.3s
Its build time seem to be around 7s/4.5s/0.8s (just the
dpkg-buildpackage part).
>>
>> Running Lintian on all our t/tests artifacts[3], this patch reduces the
>> user time with about 6 minutes. That is a drop from ~21 to ~15 minutes
>> in time's output. Real time seems to have dropped to 5m 30s from 6m
>> {0..30}s.
>>
>
> This seems to have dropped to (still at -9):
>
> [...]
2m31.5s / 1m29.2s / 1m31.3s
This was done at master being f14d0f7f, but as I recall the timings
haven't changed at lot since fa546e3 + 76f1c80 (Jan 31st).
~Niels
--
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
