[svn.haxx.se] · SVN Dev · SVN Users · SVN Org · TSVN Dev · TSVN Users · Subclipse Dev · Subclipse Users · this month's index

Re: issue 2043 profiling data

From: Justin Erenkrantz <justin_at_erenkrantz.com>
Date: 2004-09-09 03:38:15 CEST

--On Monday, September 6, 2004 7:36 AM +0200 Erik Huelsmann
<e.huelsmann@gmx.net> wrote:

> So real and user are not that far apart. Do you mean that real is wall-clock
> and user is processor time? Tobias was trying to convince me the problem was
> with swapping and waiting for the I/O, but if you are correct, then swapping
> will not be the cause of this long execution time.

*If* gprof isn't giving bogus data (what does oprofile say if you have access
to it?) and look_up_committables is really the big bottleneck, I'd say that
the iterative strcmp case at a roughly O(300,000^2) is the problem. 300,000
entries in the array, we do a *linear* search on the array, and then call
look_up_comittables on each of the 300k files in the directory, exit at first
match. I'd think a better data structure there should remove that bottleneck.
And, this would explain why the time isn't sys time - it's CPU time doing
strcmp's. My hunch from a 30 sec glance at the code. -- justin

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Thu Sep 9 03:38:43 2004

This is an archived mail posted to the Subversion Dev mailing list.

This site is subject to the Apache Privacy Policy and the Apache Public Forum Archive Policy.