[PATCH] revlog experiment

Chris Mason mason at suse.com
Fri Mar 3 14:09:18 UTC 2006


On Friday 03 March 2006 08:50, Valentino Volonghi aka Dialtone wrote:
> On Thu, 2 Mar 2006 20:25:22 -0500, Chris Mason <mason at suse.com> wrote:
> >The main goal of posting here is for benchmarking the delta vs parent and
> >packed file changes.  In my tests here, hot cache usage is a little slower
> >and cold cache is much faster.  Much of the hot cache cost is going into
> >figuring out if a packed log should be used or not, basically I need
> >to optimize my new filelog.islog(), and improve caching of the biglogs
> >during hg checkout.
>
> Another measurement:
> I checked out http://hg.fr.freebsd.org/ports/ and I converted the metadata
> using hg + your patch.
>
> with your patch:
> dialtone at aiolia /Volumes/dati/Sviluppo/ports-revlogng$ du -d 1 -h
> 418M    ./.hg
>
> dialtone at aiolia /Volumes/dati/Sviluppo/ports$ du -d 1 -h
> 1.2G    ./.hg
>
> Seems quite a lot better in space.

Wow, that's much better than I expected.  What .hgrc parameters did you use 
for the manifest/changelog?  How big is the manifest?

Could I trouble you to benchmark the cold cache checkout times? (just hg 
update -C tip with out any files currently checked out).

-chris







More information about the Mercurial mailing list