Newsgroups: comp.parallel.pvm
From: easchwei@eos.ncsu.edu
Subject: PVM 3.3.[78] on RS/6000 problem
Keywords: Help me, I'm going insane
Organization: North Carolina State University
Date: 2 Oct 1995 20:10:09 -0500
Message-ID: <9510030109.AA03496@c11013-212dan.ece.ncsu.edu>

I'm having a really strange problem in trying to get a PVM
application written in Fortran 77 to run under AIX 3.2 on 
an RS/6000.  The same code runs on a DECstation with the 
same PVM (3.3.8) OK.  I also had it running on SunOS, but
they've improperly installed Solaris over top of that machine
this past weekend.

I did some experimentation and found that if I take out most
of the procedures from the Fortran code and compile to get
a much, much smaller binary the program will spawn copies
of itself, do nothing (I took out the code), and terminate
OK.  However, if I just comment out the meat of the program
leaving only the set-up and spawn calls, the program mysteriously
bombs.

The unmodified version of the original program was creating
literally many hundreds of 0 byte files in /tmp until it had 
exhausted the filesystem's inodes and then died.  The modified
code, which does nothing but spawn several copies of itself and 
then shutdown, results in a segmentation fault.  The exact same
code with the subroutines not linked runs fine!  So, it appears 
that the problem is somehow related to the executable's size.

Does anybody know about any "weird" feechurs for PVM on AIX that
could be shooting me?  Any help or insight would be greatly 
appreciated.

I've tried linking to both PVM 3.3.8 and 3.3.7 with the same
bizarre results.

--
Eric

