Newsgroups: comp.parallel.mpi
From: Christopher Ditchman <cjd@yosemite.jpl.nasa.gov>
Reply-To: cjd@yosemite.jpl.nasa.gov
Subject: Memory Leak in MPI for Cray Release 1.4a?
Organization: Jet Propulsion Laboratory
Date: Wed, 10 Jul 1996 15:31:24 +0000
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Message-ID: <31E3CCC6.109F@yosemite.jpl.nasa.gov>

Hello,

I was wondering if anyone has heard of / can confirm the existence
of a memory leak existing in MPI for the Cray T3D (release 1.4a)
involving derived datatypes not being freed properly.  

The Cray consultants on lab seem to agree that the code I am
using is correct and that a memroy leak is within MPI.  The
main program is written in FORTRAN 77 and makes use of 
a genetic algorithm library (which uses MPI) written in C.

Do later versions of MPI fix the memory leak?  Are there work
arounds for this sort of problem?  Right now I am faced with
the prospect translating all of the FORTRAN to C in order to
change all variables from 64 to 32 bits to forstall the 
inevitable memory failure.  I am also considering switching to
PVM.  Ideally, I would like to retain as much of the FORTRAN as
possible.

Thanks in advance,

Chris

-----
Christopher J. Ditchman		Tracking Systems and Applications Group
cjd@yosemite.jpl.nasa.gov	Jet Propulsion Laboratory
						4800 Oak Grove Dr.
						Pasadena, CA 91109 MS 238-600

