Newsgroups: comp.parallel.mpi
From: dinucci@win119.nas.nasa.gov (David C. DiNucci)
Subject: Re: MPI on Shared Memory Systems
Organization: NAS - NASA Ames Research Center, Moffett Field, CA
Date: 29 Apr 1996 18:47:15 GMT
Message-ID: <4m32rj$b34@cnn.nas.nasa.gov>

In article <4m1bd7$jfs@murrow.corp.sgi.com> salo@mrjones.engr.sgi.com (Eric Salo) writes:
>> 	I am trying to implement and MPI program on the SGI power challenge
>> array. What I would like to do is to have some variables common to all
>> the processes. I dont want to explicitly pass messages as these variable are
>> frequently updated. I would like to have them as shared memory locations
>> accessible to all the processors. Is this possible using MPI? If so, how?
>
>This is a rather nasty area. As soon as you teach your application to use
>shared variables, you will destroy your portability to any other MPI
>implementation. One might also logically ask at this point why bother with
>MPI at all once you have this capability in place.

You may want to take a look at some work I've been doing in this area,
called the Cooperative Data Sharing (CDS) System.  See

  http://www.nas.nasa.gov/NAS/Tools/Projects/CDS

The goal (which has been at least partially achieved) is to provide a single
user interface for communication which will not copy on shared-memory systems
and will still work efficiently on message-passing systems.  The result is
something you might call "virtual shared messages".

The interface right now (CDS1) may be too low level for many users.  CDS2, a
higher level interface, is in the works, and will be more like MPI.  The MPI
committee has made some mention of adding some functionality like this to MPI
someday.

 -Dave
-- 
===============================================================================
David C. DiNucci    | MRJ, Inc., Rsrch Scntst  |USMail: NASA Ames Rsrch Ctr
dinucci@nas.nasa.gov| NAS (Num. Aerospace Sim.)|        M/S T27A-2
(415)604-4430       | Parallel Tools Group     |        Moffett Field, CA 94035

