[p4] Fwd: Trying to improve performance

[Chris Weiss]

Sorry, should have CC'd the list on this.

---------- Forwarded message ----------
From: Chris Weiss <chris.weiss at gmail.com>
Date: Jun 22, 2006 9:17 AM
Subject: Re: [p4] Trying to improve performance
To: Neal Firth <NealF at sageright.com>


We go the proxy server route as well, which helps us as the bulk of
our network is still only 100mb. The proxy is located on the same
small 1gb switch as the build machines. Our timing tests showed that
syncing across the 1gb switch was actually faster than syncing with
the proxy on the same box that's syncing (all of our build machines
and the proxy are mid-level Dell workstations w/PATA disks).

Half of our builds are done from scratch, so the savings are substantial.

On 6/21/06, Neal Firth <NealF at sageright.com> wrote:
> > The second part of that is when the build servers detect a change,
> > they all try to sync at the same time and there they really bring
> > Perforce down to its knees. Is there a way to avoid that other than to
> > stagger the syncs (which would increase the time until we know if the
> > build succeeded or failed)?
>
> Sounds like either a lot of files are changing between builds or you're
> forcing a "clean" environment prior to every build.
>
> You might consider having your build servers talk to a separate P4 Proxy
> server. There is still general server overhead to make sure that the proxy
> is up to date - but it's a lot less than dealing with the individual build
> servers. I've also heard of using a proxy on each build server - oh the joys
> of big disks. More general server overhead, but very fast build workspace
> population.
>
> With a little router/ bridge/ NIC magic you can arrange it so that the build
> servers only talk to the proxy server and don't add load to the general
> developer network.
>
> Various combinations of these techniques have worked very well for several
> of my clients with similar problems.
>
> Regards,
>
> // Neal - PCP/ PCT
> _______________________________________________
> perforce-user mailing list  -  perforce-user at perforce.com
> http://maillist.perforce.com/mailman/listinfo/perforce-user
>


--
-Chris


-- 
-Chris