Newsgroups: comp.parallel.pvm
From: James Arthur Kohl <kohl@msr.epm.ornl.gov>
Subject: Re: xpvm
Organization: Oak Ridge National Lab, Oak Ridge, TN
Date: 27 Jul 1995 15:32:26 GMT
Message-ID: <3v8bia$ker@stc06.ctd.ornl.gov>

dailey@iastate.edu (Lyle D Dailey) wrote:
>
> 
> 
> Does anyone know if xpvm will work with a SPMD PVM code
> using parent-child approach?  I've installed xpvm,
> and it works fine with a simple master-slave code I have,
> but it doesn't want to work with a parent-child code.
> 
> Is it a known deficiency of xpvm, or should I continue searching
> for the real problem?

Hi Lyle,

The same "problem" exists with the regular PVM console.  It's not
so much a problem with XPVM/PVM but with the logic of some SPMD
programs that use pvm_parent() to determine who the first worker
task is.  If the task happened to be spawned from the shell then
that works just fine, but otherwise either XPVM or the console is
the task's parent.

The best way to figure out who should spawn the remaining SPMD workers
is to use the PVM group server "pvmgs".  You can make a call to
pvm_joingroup() with an arbitrary group name, and then check the
instance number returned.  If you are instance zero, then you are
the first and should spawn the rest.

Hope this helps,

All the Best,

	Jim
(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:

    James Arthur Kohl, Ph.D.                "Da Blooos Brathas?!  They
    Oak Ridge National Laboratory               still owe you money, Fool!"
    kohl@msr.epm.ornl.gov
    http://www.epm.ornl.gov/~kohl/          Long Live Curtis Blues!!!

:):):):):):):):):):):):):):):):):):):):):):):):):):):):):):):):):):):):):):):)


