Newsgroups: comp.parallel.pvm
From: rex@cabsparc.larc.nasa.gov (Rex Min)
Subject: Re: PVM and the Paragon
Organization: NASA Langley Research Center, Hampton, VA  USA
Date: 1 Jul 1994 12:15:38 GMT
Message-ID: <2v11da$20p@reznor.larc.nasa.gov>


In article <2v0o32$2dim@info2.rus.uni-stuttgart.de>, smiller@intelsun.rus.uni-stuttgart.de (Shannon Miller) writes:
|> Steve Heistand (heistand@iastate.edu) wrote:
|> : I have oompiled and gotten PVM to seemingly work on the Paragon, however
|> : I cant seem to get it to use more than one node.  Anyone else
|> : get PVM to work?
|> : steve
|> 
|> Yes, it works.  We've been using it here for the better part of a year
|> now.  You can run PVM on many Paragon nodes, but you can't add the
|> Paragon to a "network of PVM systems."
|> 
|> --Shannon

Perhaps the problem is that your PVM processes are spawning on the
service nodes, rather than the compute nodes? I believe that 
all PVM processes spawned from UNIX prompts will wind up in the
service partition (even if you use a -pn flag, etc!). 

Try writing a quick "spawner" process whose job is nothing more
than to spawn the real working processes. If you avoid PvmMppFront,
the spawner (running on the service node) should spawn tasks in the
compute partition.

As for the "network of PVM systems," if that means a Paragon
combined with other workstations (like Sparcs), I may have gotten it
to work. The key is to start up PVM on the Paragon FIRST, and
then add the other workstation hosts. I'm looking into a few
possible bugs with this setup, though.

-- Rex

******************************************************************************
  Rex K. Min                               Any and all opinions in this
  c/o NASA Langley Research Center         message are mine; don't go 
  rex@cabsparc.larc.nasa.gov               blaming good 'ol NASA!

