From: salo@cray.com (Eric Salo)
Newsgroups: comp.parallel.mpi
Subject: Re: MPI_ABORT on SGI O2K
Date: 9 Sep 1998 16:10:21 GMT
Organization: Silicon Graphics, Inc.
Message-Id: <6t699d$rvb$1@walter-fddi.cray.com>
References: <6t5353$nic$1@fir.prod.itd.earthlink.net>


> My code uses MPI_ABORT to stop all processors in case any hit a
> fatal error.  It works fine on an IBM SP and a SUN HPC machine, but
> on the SGI Origin, I get the error message: "MPI_FINALIZE() not called"
> and a bunch of core dumps from the nodes.

We should probably clean this up. The problem is that we aren't currently
treating MPI_Abort() as a special case, so the usual tests for abnormal
termination aren't being turned off.

You could use the PR_COREPID option to prctl() to eliminate all but one
of the core files, but that's about all I can think of at the moment.

--
Eric Salo      Silicon Graphics      salo@sgi.com