Verification of a working pvm3 setup

[Robert G. Brown]

On Tue, 27 Jun 2000, Bob P wrote:

> I have a 2 node linux beowulf prototype.  I have
> loaded pvm3 on each on an nfs mounted directory.  PVM
> loads fine on the master and can add the second node
> with no problem, however, I have attempted to try a
> sample master/slave app with no luck.  I run the
> master program but it just enters an endless loop.  I
> modified the slave so that it doesn't do calculations,
> but just returns a float to the master.  Both apps
> compiled without a problem, and the master shows up on
> the pvm console with a 'ps -a' command.  Does anyone
> have any ideas that might get me past this problem?

Check your path(s).  PVM requires that the slave binary live in one of a
few very specific places in order to be able to successfully spawn it.
For a private user, for example, this might be ~/pvm3/bin/LINUX/ on a
linux system.

Beyond that check out the contents of e.g. /tmp/pvmX.log to see what pvm
says when it tries to spawn the slave.  You can sometimes get some help
from running at test under xpvm -- it provides you with visual access to
a lot of the logfile's running contents.  However be warned -- an
application that is "dense" in pvm activity can put so much pressure on
xpvm (which basically sucks up memory to buffer the I/O channel) that it
can exhaust swap VERY QUICKLY and crash your system.  Really xpvm could
use some judicious hacking putting some sort of memory limits on it as
it can literally run away with your system before you can respond...

   Hope this helps (as always).  If it doesn't, try to snoop a bit more
and provide more detail about what is going on on the next pass.

   rgb

Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at phy.duke.edu