Newsgroups: comp.parallel.pvm
From: Peter Lister <p.lister@cranfield.ac.uk>
Subject: Re: Restricting PVM
Organization: Computer Centre, Cranfield University
Date: Mon, 15 Jul 1996 10:22:50 +0100
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Message-ID: <31EA0DEA.6D96D174@cranfield.ac.uk>

> I just recently installed the PVM version of POV-Ray, a
> freeware raytracer and it works fine. When I start the
> PVM console I do it from my machine and then start up the
> other hosts on other machines. However these other machines
> may be in use by other users and when I do a raytracing,
> I don't want to bog their machine down while they are
> one it.
> 
> Is there a way that the user on one of the machines that is
> running the pvmd process can instruct the pvm console to
> NOT use their machine? Right not I ensure they are not on
> but is there anything they can do short of killing the
> pvmd process manually to temporarily remove their machine
> from the list of available hosts?

I wish. For as long as PVM, MPI etc continue to rely on rsh to start
remote processes, then chaos will reign. What we both want is for
interactive, traditional batch and "parallel" PVM style user processes
to be controlled by a common resource allocation system.

The GNQS project is heading the right direction, in that it is moving
batch queueing towards a well defined protocol. There is no reason that
I can see why PVM couldn't be an NQS client. Initially, just starting
and stopping remote pvmds using the local batch queueing system would be
an advance, but ultimately why the two styles of working should merge
into a common environment for compute intensive work.

Is anyone in PVM community working on this? Is anyone ever likely to?

