Re: FreeBSD CVS Question

On Wed, 14 Jan 2004, Richard Coleman wrote:

> > FYI, I've often found that tunneling CVS over SSH is substantially faster
> > than NFS.  This appears to be a property of CVS doing some sort of
> > explicit pipelining, whereas with NFS, it spends a lot of time blocked on
> > synchronous stat() operations against the CVS repository.  Of course, last
> > time I used CVS over NFS seriously was on 10mbps ethernet, so this well be
> > a non-event with gigabit.  :-)  Setting up the SSH connection is slightly
> > more expensive, but the reduced apparent latency makes a big difference. 
> 
> If you are using NFS, why do a cvs checkout on the clients at all?  Just
> do the checkout once (on the NFS master) and mount /usr/src on the
> clients.  I'm doing this for both src and ports using the procedures
> outlined in the development(7) man page. 

Ah, because I maintain lots of systems with slightly different source
revisions, so I have a checkout per machine, and build on each machine. 
If they were all running the same revision, I'd build all on the central
box, and then make installworld/etc over NFS. 

> But I do agree with Robert that if you want to do an actual cvs
> checkout, use ssh instead.  Once you've created ssh keys and defined
> CVS_RSH, it's just as convenient as NFS and faster. 

In fact, with recent versions of CVS on FreeBSD, you don't even need to
set CVS_RSH -- we default to ssh rather than rsh. 

Robert N M Watson             FreeBSD Core Team, TrustedBSD Projects
robert_at_fledge.watson.org      Senior Research Scientist, McAfee Research