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

RE: Re: checkout performance

From: Juergen Richtsfeld <juergenr_at_segue.at>
Date: 2005-12-21 08:40:47 CET

thanks for your tip, i would also like mod_auth_kerb, but my recent testings were made without security things at all. but i thank you very much for your reply as it shows me, not to use mod_auth_kerb at the moment.

jürgen

> -----Original Message-----
> From: Tony Butt [mailto:tony.butt@cea.com.au]
> Sent: Wednesday, December 21, 2005 12:09 AM
> To: users@subversion.tigris.org
> Subject: Re: checkout performance
>
> We run our server on a 2.8HGz Xeon processor running SuSE Enterprise
> Linux 9, running apache2.0.49 and subversion 1.2.3.
>
> Our authentication system uses mod_auth_kerb to refer to a
> Windows 2000
> domain controller, and authz_svn to control access to various parts of
> the repository. There seems to be a bottleneck with the interaction
> between mod_auth_krb and mod_authz_svn, where the authentication
> information is not a cached, but checked for EACH FILE in the
> repository
> that a transaction looks at.
>
> I found that something like 4 DNS lookups were also being performed
> for each file, and fixed most of that by hardcoding IP addresses into
> the krb.conf configuration file. This has reduced the server CPU load,
> but checkouts and log operations are still much slower than we would
> like, and something like 10x slower than using svn: protocol, with no
> authz_svn (and no auth_kerb). We like mod_auth_kerb for the
> single sign
> on ability it gives us, but the lack of credential caching is really
> killing our performance. We have experimented with bdb and fsfs
> backends, but found no definitive performance differences, so
> you might
> be able to gain a little there, but not much.
>
> In all likelihood, your problems are related to authentication and
> authorisation - maybe mod_auth_ldap would be better for you?
>
> Tony Butt
>
> CEA Technologies
> Canberra, Australia
>
> Res Pons wrote:
> > Juergen,
> >
> > To make you feel better: I serve Subversion on a super fast Linux
> > Enterprise server with 4 Xeon processors, tons of disk
> space, and tons
> > of RAM. All the users' desktop clients are Pentium4 @3-4GHZ
> with WinXP
> > + 512MB to 1GB of RAM. And a complete fresh check out still takes
> > 20-25 minutes for a 250MB project! Sometimes my patience just runs
> > out. Not to mention it's extremely CPU intensive and takes my PC
> > hostage. I literally cannot do anything until SVN is done checking
> > out. NOW, consider yourself lucky :)
> >
> >
> >
> >
> > ----Original Message Follows----
> > From: Kjell H Andersen <kha-svn@lentic.net>
> > To: Ryan Schmidt <subversion-2006Q1@ryandesign.com>
> > CC: Juergen Richtsfeld <juergenr@segue.at>,
> users@subversion.tigris.org
> > Subject: Re: checkout performance
> > Date: Mon, 19 Dec 2005 15:08:27 +0100
> > MIME-Version: 1.0
> > Received: from tigris.org ([64.125.133.100]) by
> > bay0-mc3-f3.bay0.hotmail.com with Microsoft SMTPSVC(6.0.3790.211);
> > Mon, 19 Dec 2005 06:15:35 -0800
> > Received: (qmail 13122 invoked by uid 5000); 19 Dec 2005
> 14:09:57 -0000
> > Received: (qmail 12974 invoked from network); 19 Dec 2005
> 14:09:37 -0000
> > X-Message-Info: JGTYoYF78jFSuACWoi/6zybCOlS8nnQQqeMXsbFvV/Y=
> > Mailing-List: contact users-help@subversion.tigris.org; run by ezmlm
> > Precedence: bulk
> > X-No-Archive: yes
> > list-help: <mailto:users-help@subversion.tigris.org>
> > list-unsubscribe: <mailto:users-unsubscribe@subversion.tigris.org>
> > list-post: <mailto:users@subversion.tigris.org>
> > Delivered-To: mailing list users@subversion.tigris.org
> > User-Agent: Thunderbird 1.5 (Windows/20051025)
> > References:
> <3145C5555D21AD408CFA3A4BFAC77B976B0082@atmail1.segue.com>
> > <593D3588-D352-4F3D-8770-65B98B3C2B40@ryandesign.com>
> > Return-Path:
> users-return-42597-pons32=hotmail.com@subversion.tigris.org
> > X-OriginalArrivalTime: 19 Dec 2005 14:15:36.0035 (UTC)
> > FILETIME=[AE33CB30:01C604A6]
> >
> > I've had some experience with large check-out/in. The problem isn't
> > the size itself, but the number of files.It looks like most file
> > systems has a noticeable overhead for each file it has to
> create. I've
> > had some problems related to this when checking out a
> working copy to
> > a nfs mount. NFS performs really poor when trying to perform
> > operations on many files.
> >
> > But to answer your question, I don't think 10-12 minutes
> are that much
> > considering the number of files. Remember that all later operations
> > will only be performed on a small amount of data. If you keep your
> > working copy on a network drive, you can probably improve
> performance
> > by moving it to your local drive, but that might lead to
> some backup
> > issues..
> >
> > Kjell
> >
> >
> > Ryan Schmidt wrote:
> >> On Dec 19, 2005, at 13:18, Juergen Richtsfeld wrote:
> >>
> >>> I'm just about to try around with subversion. But it seems to be
> >>> quite slow.
> >>>
> >>> my server has about 1GHZ, 1GB ram, windows 2000 server.
> >>> my repository has about 350 MB in about 11000 files. my
> repository
> >>> was created with default settings (svn 1.2) (bdb). my
> client is a P4
> >>> 3Ghz with 1 gig ram.
> >>>
> >>> doing a clean checkout takes about 12 minutes with apache
> and about
> >>> 10 minutes with svnserve (100 mbit lan).
> >>>
> >>> I'm currently testing with fsfs. are my measurements
> normal? i think
> >>> it's quite slow. the problem is, we're currently using ms
> >>> sourcesafe, which is much faster (but lacks lot of features).
> >>
> >> I don't know if 10 or 12 minutes is normal for a working
> copy of that
> >> size. I have heard that checkouts on Windows take longer than on
> >> other operating systems. But the important thing to
> remember is that
> >> you only do a full checkout very rarely, usually just when
> you first
> >> start working with Subversion. From then on you can just
> update the
> >> existing working copy, which should be much faster.
> >>
> >>
> >>
> >>
> ---------------------------------------------------------------------
> >> To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
> >> For additional commands, e-mail: users-help@subversion.tigris.org
> >>
> >>
> >
> >
> >
> ---------------------------------------------------------------------
> > To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
> > For additional commands, e-mail: users-help@subversion.tigris.org
> >
> > _________________________________________________________________
> > FREE pop-up blocking with the new MSN Toolbar - get it now!
> > http://toolbar.msn.click-url.com/go/onm00200415ave/direct/01/
> >
> >
> >
> ---------------------------------------------------------------------
> > To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
> > For additional commands, e-mail: users-help@subversion.tigris.org
> >
>
>
>

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Received on Sun Dec 25 01:04:56 2005

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

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