See inline below.
-----Original Message-----
From: Nico Kadel-Garcia [mailto:nkadel_at_gmail.com]
Sent: Tuesday, August 17, 2010 11:36 AM
To: Tami Doyen
Cc: users_at_subversion.apache.org
Subject: Re: SVN server timeouts
On Tue, Aug 17, 2010 at 1:30 PM, Tami Doyen <Tami_Doyen_at_pmc-sierra.com> wrote:
> We are running 1.6.11 server and client. The server is Red Hat EL5. Our
> repositories are used for chip development and are huge with multi-gig files
> sizes.
And they're binary and filled with thousands of files? And don't tell
me, your clients are writing to CIFS filesystems?
]] TD - Not CIFS, most clients are Unix, NFS
I've recently had to
deal with this with multiple Gig repositories, admittedly with smaller
files.
May I suggest first, making sure that your checkouts occur to local
filesystems, not CIFS shares on Windows? It seems to make a huge
difference with my large repositories with many thousands of files,
some of them hundreds of MBytes. Writing to local disk, or to an NFS
mounted directory, makes a huge performance improvement.
]] TD - yes, our software teams are using Windows and that is much slower, but these users are checking out locally to NFS
Other approaches are sensitive to the amount of churn in the modified
files. A local replicated server, or working copy, can be very helpful
to getting that first working copy locally and updating only as
needed.
]] TD - Yes we are replicating to local servers first
> The challenges with moving huge amounts of data back and forth between
> remote worldwide sites is causing severe delays. There is plenty of
> information about hosting/replicating SVN servers, but there is little
> information out there about some of the difficulties when dealing with
> multi-gig files. Most information is focused on source code management and
> doesn't deal with files that are more than a few MB.
That's partly because you're not talking about source, you're talking
about what are effectively binary images in the multi Gigabyte file
size range. It's a whole different set of issues.
> We would also like to get in touch with an organization that is working with
> the obstacles we are, mostly due to us using svn for chip development. We
> also use svn for software development without any issues this severe.
Good luck with that: Gibabyte files constitute their own issue. Would
you be helped by a proxy server, which could cache the bulky files in
a local repository for you without having to tweak your clients or
other configurations? Or can you, procedurally, copy a local working
copy and "svn update" that copy?
]] TD - We are also caching the data and it has improved performance but not well enough. Thanks for the suggestions!
Received on 2010-08-17 20:50:06 CEST