Newsgroups: comp.parallel.pvm
From: jeff@adeno.wistar.upenn.edu (Jeffrey Taylor)
Subject: Re: How can I spawn only one process on one machine?
Organization: University of Pennsylvania
Date: 8 Jan 1997 19:11:40 GMT
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Message-ID: <5b0rhc$4sf@netnews.upenn.edu>

Ok, I have two thoughts here.  First, it would be easy to spawn the new
job specifically on the host that just returned a result using the 
PvmTaskHost flag in pvm_spawn.  
	Second, are you sure you want to do this anyway?  The overhead of 
spawning a new process is substantial and most program models intially 
spawn a slave on every node which remain active for the entire run time 
and then don't start new slaves once the computation has begun.  Starting
and stopping slave durring the computation is sure to be a large 
performance drain.

If you must spawn new slaves continually, you'll just have to be carefull
about allocating memory.  If your malloc call fails, have your slave shut
down gracefully and then respawn.  Unfortunately, it may bounce around your
virtual machine forever!



Jeff Taylor

-- 
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Jeffrey S. Taylor			        Molecular Modelling	
The Wistar Institute and			Software development
The University of Pennsylvania			Molecular Biology
Philadelphia, PA 19104          E-Mail    Jeff@aries.wistar.upenn.edu	
(215) 898-2202			WWW     http://adeno.wistar.upenn.edu