Re: speed: pserver vs mount of repository

[Mark D. Baushke]

Richard Pfeiffer <address@hidden> writes:

> I know it's not advised to NFS mount the cvs
> repository to the machine on which the cvs binary
> resides.  However, we have a user group that is
> convinced we have to do so for speed reasons. 
> (Doing updates of a massive repository approx.
> every 25 minutes)
> Would anyone happen to know of any test
> comparison cases (pserver connection vs actual
> mount) regarding this or have any opinions on the
> subject?

The correct question is this:

  "Is it ever acceptable that a 'cvs commit' may corrupt the repository
  without any notification of any problems whatsoever until much time
  has passed and there is a need to checkout an old version of the
  repository that is found to be corrupted due to earlier NFS usage?"

If the answer is: 

  "We do not care about old versions or if they might happen to be
  corrupt.", then by all means feel free to use NFS to do the checkouts.

If the answer is: 

  "Correctness is more important than speed."

then just say no to using the NFS approach.

By the way, you may find that having a high bandwidth connection between
your cvs pserver machine and the datastore for the repository (hopefully it
is in some kind of UFS filesystem rather than NFS) will benefit from using
the -zN (where N is one of 1,2,3,...,9) compression. Or, if you are using
'CVS_RSH=ssh' then you have compression and security of the packets.

I would suggest that the only test comparisons you should consider are
those you setup and run for yourself on your own hardware and networking
equipment as there are a great many factors that can reduce throughput.

        -- Mark