"Simon Knight" <Simon.Knight@itsq.com.au> writes:
> Here are the relevant sections of http.conf
> LoadModule dav_svn_module modules/mod_dav_svn.so
> LoadModule authz_svn_module modules/mod_authz_svn.so
> LoadModule php4_module modules/libphp4.so
>
> ScriptAlias /cgi-bin/ "/usr/local/apache2/cgi-bin/"
> ScriptAlias /view "/usr/local/viewcvs-1.0-dev/www/cgi/viewcvs.cgi"
>
> There is no authentication on the http side because it is read only. We
> are using a password file on the svnserve side.
>
> The speed issue was with viewing using viewcvs. We have some deep trees
> and when clicking on one of these it took in excess of a minute to
> display the next level. Running 'top' showed several httpd processes
> running at max cpu. Unfortunately I rolled back before testing the raw
> DAV interface as I had developers waiting for access, so I don't know if
> that was slow also. I did a quick 'svn status' on the root of my working
> copy and it was also slower than the older release, but is not a show
> stopper like the browsing.
Thank you, this is very important information.
ViewCVS is a separate piece of software, that makes some calls into
Subversion to do its work. So if ViewCVS browsing has become slower,
please express that by saying "ViewCVS browsing has become slower",
not "The DAV performance is so bad that it is virtually unusable and
we have rolled back to 1.0.3" :-). The latter makes assumptions about
the way ViewCVS works -- for example, that it uses DAV to access the
repository, which it doesn't. It also implies that the horrible
slowness was seen by users of a Subversion client in a working copy
(since that's what uses DAV), but this apparently wasn't the case
either.
Please always report the exact symptoms, instead of reporting theories
about their causes. Otherwise people reading your posts will get
confused about what actually happened and misdiagnose something.
> I noticed that one difference between the versions is that the older
> version is dynamically linked, while the later one is statically linked.
> This means that the two are probably not running the same versions of
> libraries, particularly neon since it is include in the distro at a
> later version than the installed one.
You're talking about the server side, right? Any idea why would one
be dynamically linked and another statically? Were they compiled
differently on purpose?
(Btw, the server side doesn't use Neon at all, only the client, and
then only if you're using the http:// access method.)
In 1.1.0, we know some operations are slower over http://, if you are
using authentication and authorization modules, such as mod_authz_svn.
The slowdown is due to security fixes. However, we do not expect
showstopper slowdowns. I don't *think* the fixes should affect
ViewCVS at all, either, though am not 100% sure about that.
You said earlier you saw a 2x slowdown with 'svnserve', but still
haven't said which operations, how much data, etc.
If there is no authentication on the httpd side, then why are you
loading mod_authz_svn, which can only work if there is authentication?
We might be able to get to the bottom of this, but we'd need a lot
more facts, and that might mean you'd have to upgrade back to 1.1 rc4
to gather those facts. Don't know if that's worthwhile for you. Your
call, obviously.
Best,
-Karl
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Received on Fri Oct 1 05:56:10 2004