On Sun, Feb 15, 2004 at 06:36:36PM -0600, kfogel@collab.net wrote:
> > * I don't think we have a story for maintaining vendor branches (a
> > la successive "cvs import" commands) which works for a large tree.
> > svn_load_dirs.pl is too slow, and simply importing successive
> > versions into different trees gets you no historical association
> > between the files.
>
> Do we know whether this is because svn_load_dirs.pl isn't implemented
> as efficiently as it could be? Not implying anything; I really don't
> know, haven't looked at it carefully.
>
> If they quantified "too slow", that would be handy data to have.
It's obviously because of wc being used in svn_load_dirs.pl.
Here is a benchmark importing perl 5.8.0 tree:
svk: 72 wallclock secs (23.66 cusr + 6.64 csys = 30.30 CPU) [1]
svn: 190 wallclock secs (39.96 cusr + 26.45 csys = 66.41 CPU)
and then perl 5.8.2 tree over the branch:
svk: 85 wallclock secs (28.45 cusr + 7.45 csys = 35.89 CPU)
svn: 298 wallclock secs (80.81 cusr + 46.89 csys = 127.70 CPU)
the size of the trees:
48M perl/perl-5.8.0
52M perl/perl-5.8.2
[1] vanilla "svn import" is as fast as fresh svk import
Cheers,
CLK
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Mon Feb 16 12:00:04 2004