Tobias Ringstrom <email@example.com> writes:
> On Tue, 8 Apr 2003, Greg Stein wrote:
> > On Tue, Apr 08, 2003 at 02:37:43PM -0700, Christian Daudt wrote:
> > > Look at your memory consumption if you're using it via http -
> > > swapping might explain the difference. httpd still grows to many
> > > hundreds of megabytes regularly when I'm using it (my simple
> > > solution was: install more memory ;-)
> > Hunh? It shouldn't do that at all.
> > What version are you using?
> > No... the speed is simply due to a *lot* of I/O on the client. We also make
> > quite a few roundtrip requests to the server, and also some needless
> > requests. These things add up.
> And don't forget the gazillion of fsync() calls on the server side as I
> reported last week. I got over 600 fsync() calls on the server for a 30
> file repository checkout. The drive sounds as if it's going to jump out
> of the case. I do not know how mush performance this steals though.
> It seems ra_svn and ra_file is *a lot* faster than ra_dav. Do you think
> ra_dav will ever be as fast as ra_svn and ra_file, or is there a
> fundamental performance problem with the dav protocol?
Using ra_dav involves marshalling data over a socket from one process
to another, it's unlikely ever to be as fast as ra_file which doesn't
have that overhead.
ra_svn is often faster than ra_dav, in part because it uses a protocol
that is custom designed to support Subversion (it doesn't suffer from
problems like issue 1161) and partly because the ra_svn server is far
simpler than Apache. The simple server is an advantage because it
makes it easier to fix problems, I looked at issue 773 some time ago
and decided I would need to learn about mod_dav/httpd to have much
chance of fixing it.
The ra_svn server has some serious limitations compared to the ra_dav
server, particularly on a Windows machine. It's probably true to say
that the ra_svn server provides fewer features than the ra_dav server.
It's possible that in the long run the ra_dav server will scale better
than the ra_svn server, one of the arguments for using Apache is that
it provides a sophisticated server, although scaling is academic while
the base performance remains poor.
To unsubscribe, e-mail: firstname.lastname@example.org
For additional commands, e-mail: email@example.com
Received on Wed Apr 9 02:06:28 2003