Newsgroups: comp.parallel.pvm
From: andy@henry.jpl.nasa.gov (Andy Boden)
Subject: Q: Manipulating PVM default process mapping
Organization: Jet Propulsion Laboratory (California Institute of Technology)
Date: 6 Sep 1994 17:49:41 GMT
Message-ID: <34ia3m$69v@elroy.jpl.nasa.gov>

Okay, I'll try again.

For sake of simplicity I'd like to spawn N slave tasks onto M nodes in
a PVM configuration with (something like) the following (single)
command:

    if ((info = pvm_spawn(SLAVE_NAME,(char**)0,
			  PvmTaskDefault,"",
	                  num_slaves,slave_ids)) != num_slaves)
      (print out an error message)

From my own (limited) experience this command results in a
more-or-less uniform distribution of tasks across nodes.  In
particular, this uniform distribution of tasks seems to be completely
independent of the speed settings of the constituent machines from the
PVM hostfile: machines with comparitively slow speed settings have the
same expected number of processes as machines with high speed
settings.

So my question is: is there any known way to effect the default task
placement made with the pvm_spawn command?  A related question that
comes to mind is if the speed setting isn't used in the default task
placement algorithm, in what context is it used?

(BTW -- please don't suggest that I simply start-up the desired number
of tasks on each machine individually -- of course that will work --
but requires the maintenance of a machine database that I would prefer
to avoid.)

Thank you for your time in advance.

							- A.B.

 --


"...it doesn't matter how beautiful your theory is, it doesn't matter
how smart you are -- if it doesn't agree with experiment, it's wrong."

						- R.P. Feynman

