On Feb 27, 2005, at 5:47 AM, Greg Hudson wrote:
> If you've been following
> http://svn.haxx.se/dev/archive-2005-02/0897.shtml ("ugly problem found
> while trying to test KDE SVN"), then you've seen me conclude that our
> auto-merging code can be ruinously inefficient on big repositories.
> Although Eric has fixed the auto-merge memory leaks in FSFS (his
> change will need to be back-ported to BDB), doing multiple complete
> predecessor walks in a repository like kde-i18n is prohibitively slow.
>
Greg, thanks not only for doing the legwork in identifying the problem,
but for proposing a new algorithm. I have to admit that I'm really
surprised that such a large performance/scalability bug exists in such
a fundamental algorithm right at the core of our commit logic... I'm
equally surprised that we've never noticed it. I remember Karl and
Mike spending a good couple of days (back in early 2001) discussing and
carefully implementing the current algorithm, so maybe they'll be able
to comment on those questions you asked.
Wow. Has an issue been filed? This seems pretty high-priority to me.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Sun Feb 27 15:44:01 2005