Newsgroups: comp.parallel.pvm
From: Graham Edward Fagg <sssfagg@csres.cs.reading.ac.uk>
Subject: Re: Location of spawned tasks/group functions
Organization: University of Reading, U.K.
Date: Thu, 23 Feb 1995 12:43:12 +0000
Mime-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Message-ID: <Pine.SOL.3.91.950223123611.11610A-100000@suma3>

On 20 Feb 1995, Niall Adams wrote:
> Hello,	

>         I have a program which consists of a master program that 
> spawns p slaves. I notice that a slave task is often spawned on the
> local host, where the master task is executing. It seems to me that
> this must slow the execution time of the PVM program since the local
> host is doing more work than is desirable. Is this correct? If so,
> is there a way to force slaves to be spawned on machines other than
> the local host?

PVM uses a round robin system to allocate hosts to run task on. So if you 
have 8 hosts, 1 master and 8 slaves then one of your machine is going to 
get the master and a slave...

> My slave tasks, (and indeed the master ) perform very similar 
> tasks. Is there any benefit in using 'process groups'? Are there 
> any references to their use? 

The group functions in their simple form allows you to talk to many 
processes without having to pass arrays of Id's around. Their real 
attract and advantages become clear when you start using the more complex 
calls such as barrier and esp. the pvm_reduce call with user defined 
functions. They also make it easier it interface PVM to CASE tools.

> Niall Adams
> n.adams@livjm.ac.uk 

Graham.
===============================================================================
Graham Edward Fagg ||| *** Cluster Computing Lab. ***  ||| e-mail me some time
 Computer Science ||  Software Engineering Subject Grp.  ||  G.E.Fagg@rdg.ac.uk
0734-875123 7626 | http://www.cs.reading.ac.uk/people/gef/ | PVM/MPI/LINDA/VMD
===============================================================================


