Newsgroups: comp.parallel.pvm
From: cfreese@super.org (Craig F. Reese)
Subject: xpvm/CM-5?
Summary: should xpvm work with CM-5 tasks?
Organization: Supercomputing Research Center (Bowie, MD)
Date: Mon, 20 Jun 1994 23:02:36 GMT
Message-ID: <1994Jun20.230236.27979@super.org>


xpvm/CM-5 question...

Is anyone getting xpvm to work with CM-5 tasks?  I can't even get
a simple host program to show up in the xpvm "views." 

Craig

-----More details for those who care...

What I have is a master sun that starts up a simple slave program on
other machine(s) in the configuration.  The slaves don't do much.  They
receive a message, sleep, send a message and exit.  Everything works
fine for the other machines (mostly Suns) in the configuration, but
when I spawn the slave on the CM-5 it remains "invisible" as far as
xpvm is concerned.  I do see the master Sun spawn the CM-5 task.  I
can also see it send to the task and receive from the task, but the
CM-5 end of everything remains hidden.  

The CM-5 executable is basically the same as the other Sun slaves.  It
lives in the right bin/CM5 directory and is started MppFrontEnd.  It
really doesn't use the CM-5 nodes, but I can't see where that would
really matter here.  If anything it makes life easier.

I looked at the documentation (i.e. ddpro.c line 846) and the part of
the code that sends the "newtask" trace event back from the CM-5 is
excluded due to an #ifdef.  (In fact, the CM-5, PARAGON, I860, and
SHMEM "ARCH"s are all excluded).  I looked and the xpvm trace file
doesn't seem to have _any_ messages from the CM-5 in it so I assume
that the CM-5 tasks don't send anything back.  I poked around in
some of the other source files, and I really can't tell whether it
should be sending stuff back or not.  It looks like the CP code
links with a library that's pretty similar to a standard SUN4 so
I would expect to see the trace event to come back from there; 
but I don't.

I'd be happy to just have the main host CP program generate trace
events as my application will eventually be host-node/CMMD model.
I guess I can achieve the same thing by building a CM-5 pvmd that
really only thinks it is a SUN4, but I'd like to retain the
CM-5/PVM capability for comparison/experiment purposes.

Craig

*** The opinions expressed are my own and do not necessarily reflect 
*** those of any other land dwelling mammals....

"The problem ain't what we don't know; it's what we know that just ain't so
Either we take familiar things so much for granted that we never think about 
how they originated, or we "know" too much about them to investigate closely."
-----------------
Craig F. Reese                           Email: cfreese@super.org
Institute for Defense Analyses/
Supercomputing Research Center
17100 Science Dr.
Bowie, MD  20715-4300

