You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@subversion.apache.org by David Ferguson <fe...@gmail.com> on 2007/04/02 15:51:24 UTC
Slow performance using http:// vs. svn://
Hi all,
I am working with my company's IT department to setup a Subversion server so
that we can evaluate switching from CVS to Subversion. IT would like us to
use an Apache server for repository access. However, the performance is
pretty bad compared to just running svnserve. A fresh checkout using
http:// takes 3:24 while checking out the same repository using svn:// takes
1:47. At first I suspected that Apache's LDAP authentication was slowing
things down, but I had them disable authentication with little change in the
checkout time.
Any idea why http:// would be so much slower than svn:// ? Is this
expected? Can you recommend any change to the Apache configuration to speed
things up?
thanks,
David
Re: Slow performance using http:// vs. svn://
Posted by John Waycott <ja...@cox.net>.
Kenneth Porter wrote:
> --On Wednesday, April 04, 2007 6:38 AM -0700 John Waycott
> <ja...@cox.net> wrote:
>
>> Make sure you enable compression on the Apache server. Turning it on
>> significantly improved checkout times for our setup.
>
> Do you mean mod_deflate?
>
> <http://httpd.apache.org/docs/2.0/mod/mod_deflate.html>
>
> I hadn't realized that could be combined with Subversion (or DAV). I
> thought Subversion was already compressing the traffic.
>
> It looks like mod_deflate also handles compressed input, so one could
> in principle speed up commits with this.
>
Yes - for us enabling mod_deflate made https competitive with svnserve
when doing checkouts. It has no real effect on updates from what we can
tell. As has been stated by others, it may or may not help depending on
your situation. FWIW, we are running Apache 2.0.59 and SVN 1.4.0
-- John
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Re: Slow performance using http:// vs. svn://
Posted by Mark Phippard <ma...@gmail.com>.
On 4/4/07, Kenneth Porter <sh...@sewingwitch.com> wrote:
> --On Wednesday, April 04, 2007 6:38 AM -0700 John Waycott
> <ja...@cox.net> wrote:
>
> > Make sure you enable compression on the Apache server. Turning it on
> > significantly improved checkout times for our setup.
>
> Do you mean mod_deflate?
>
> <http://httpd.apache.org/docs/2.0/mod/mod_deflate.html>
>
> I hadn't realized that could be combined with Subversion (or DAV). I
> thought Subversion was already compressing the traffic.
>
> It looks like mod_deflate also handles compressed input, so one could in
> principle speed up commits with this.
Subversion does compress the file data it sends across the wire, but
mod_deflate also allows the HTTP traffic to be compressed. Sometimes
it will speed things up sometimes it will not.
--
Thanks
Mark Phippard
http://markphip.blogspot.com/
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Re: Slow performance using http:// vs. svn://
Posted by Kenneth Porter <sh...@sewingwitch.com>.
--On Wednesday, April 04, 2007 6:38 AM -0700 John Waycott
<ja...@cox.net> wrote:
> Make sure you enable compression on the Apache server. Turning it on
> significantly improved checkout times for our setup.
Do you mean mod_deflate?
<http://httpd.apache.org/docs/2.0/mod/mod_deflate.html>
I hadn't realized that could be combined with Subversion (or DAV). I
thought Subversion was already compressing the traffic.
It looks like mod_deflate also handles compressed input, so one could in
principle speed up commits with this.
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Re: Slow performance using http:// vs. svn://
Posted by John Waycott <ja...@cox.net>.
David Ferguson wrote:
> Hi all,
>
> I am working with my company's IT department to setup a Subversion
> server so that we can evaluate switching from CVS to Subversion. IT
> would like us to use an Apache server for repository access. However,
> the performance is pretty bad compared to just running svnserve. A
> fresh checkout using http:// takes 3:24 while checking out the same
> repository using svn:// takes 1:47. At first I suspected that
> Apache's LDAP authentication was slowing things down, but I had them
> disable authentication with little change in the checkout time.
>
> Any idea why http:// would be so much slower than svn:// ? Is this
> expected? Can you recommend any change to the Apache configuration to
> speed things up?
>
> thanks,
> David
>
Hi David,
Make sure you enable compression on the Apache server. Turning it on
significantly improved checkout times for our setup.
-- John Waycott
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Re: Slow performance using http:// vs. svn://
Posted by David Ferguson <fe...@gmail.com>.
Hmm.. that makes sense that WebDAV would be slower. 2X slower is a pretty
stiff penalty though.
I have not tried the Serf client library. I didn't even know that was an
option until I read your post. From my research on the Serf web site and
the "Serf and Subversion" thread on dev@svn, it looks like Serf support made
it in to svn 1.4.3. Is Serf 0.1.0 what I should be linking with? Or do I
need the latest development version of Serf?
The libsvn_ra_serf/README states "Basic authentication" is supported. Not
sure if that includes more complicated providers like LDAP. Does anyone has
experience with this?
Has anyone done a real-world performance comparison with Serf vs. Neon?
thanks,
David
On 4/2/07, Erik Huelsmann <eh...@gmail.com> wrote:
>
> On 4/2/07, David Ferguson <fe...@gmail.com> wrote:
> > Hi all,
> >
> > I am working with my company's IT department to setup a Subversion
> server so
> > that we can evaluate switching from CVS to Subversion. IT would like us
> to
> > use an Apache server for repository access. However, the performance is
> > pretty bad compared to just running svnserve. A fresh checkout using
> > http:// takes 3:24 while checking out the same repository using svn://
> takes
> > 1:47. At first I suspected that Apache's LDAP authentication was
> slowing
> > things down, but I had them disable authentication with little change in
> the
> > checkout time.
> >
> > Any idea why http:// would be so much slower than svn:// ? Is this
> > expected? Can you recommend any change to the Apache configuration to
> speed
> > things up?
>
> The svn protocol is statefull and tailored for use by svn.
>
> The HTTP protocol is stateless and defined as the more general-use
> WebDAV protocol.
>
> So, yes, because of these reasons (and some others), there's an
> expected speed difference in favor of svn://. How big it should be or
> whether your config can be tweaked better, I don't know.
>
> I'm not sure Serf supports LDAP, but have you tried performance checks
> with the Serf HTTP client library? There should be measurable
> performance benefits from using Serf.
>
>
> bye,
>
> Erik.
>
Re: Slow performance using http:// vs. svn://
Posted by Erik Huelsmann <eh...@gmail.com>.
On 4/2/07, David Ferguson <fe...@gmail.com> wrote:
> Hi all,
>
> I am working with my company's IT department to setup a Subversion server so
> that we can evaluate switching from CVS to Subversion. IT would like us to
> use an Apache server for repository access. However, the performance is
> pretty bad compared to just running svnserve. A fresh checkout using
> http:// takes 3:24 while checking out the same repository using svn:// takes
> 1:47. At first I suspected that Apache's LDAP authentication was slowing
> things down, but I had them disable authentication with little change in the
> checkout time.
>
> Any idea why http:// would be so much slower than svn:// ? Is this
> expected? Can you recommend any change to the Apache configuration to speed
> things up?
The svn protocol is statefull and tailored for use by svn.
The HTTP protocol is stateless and defined as the more general-use
WebDAV protocol.
So, yes, because of these reasons (and some others), there's an
expected speed difference in favor of svn://. How big it should be or
whether your config can be tweaked better, I don't know.
I'm not sure Serf supports LDAP, but have you tried performance checks
with the Serf HTTP client library? There should be measurable
performance benefits from using Serf.
bye,
Erik.
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Re: Slow performance using http:// vs. svn://
Posted by David Ferguson <fe...@gmail.com>.
Yep, nothing interesting to report there. I didn't increase verbosity, but
at default levels I don't see anything unusual.
On 4/2/07, Ramanathan Muthaiah <ru...@gmail.com> wrote:
>
> > Any idea why http:// would be so much slower than svn:// ? Is this
> > expected? Can you recommend any change to the Apache configuration to
> speed
> > things up?
>
> Did you had a look at the Apache's access / error logs to find out
> what is hapenning while these checkouts are performed ?
>
> /Ram
>
Re: Slow performance using http:// vs. svn://
Posted by Ramanathan Muthaiah <ru...@gmail.com>.
> Any idea why http:// would be so much slower than svn:// ? Is this
> expected? Can you recommend any change to the Apache configuration to speed
> things up?
Did you had a look at the Apache's access / error logs to find out
what is hapenning while these checkouts are performed ?
/Ram
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org