Newsgroups: comp.parallel.mpi
From: Georg Stellner <stellner@informatik.tu-muenchen.de>
Subject: Re: MPICH amd PThreads
Organization: Technische Universitaet Muenchen
Date: Wed, 08 May 1996 19:13:53 +0200
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Message-ID: <3190D651.1CB5@informatik.tu-muenchen.de>

Lothar Thum wrote:
> 
> Hi,
> 
> now that I know that LAM/MPI isn't thread-safe, I wonder whether
> MPICH is thread-safe. I have tested MPICH with the same programs than I
> had tested the LAM implementation and I got the same results:
> 
> On one single node several threads made some communication calls
> simultanously and the program hangs up.
> If I made on each node only one communication call at a time all works
> fine and also if I saved the communication calls with mutexes.
> 
> Therefore I come to the opinion MPICH isn't thread-safe either, but
> I have heard it would.
> Is there any thread-safe MPI implementation ?
As far as I know none of the publicly available MPI implementation
is thread-safe. I even think that none of the vendor implementations
is currently thread-safe. If this has changed in the meantime, this 
would be a pretty good time for all the implementors to come up on 
the stage and comment about thread safety of their implementation.
It would also be interesting to get to know which group is really
working on a thread safe implementation. So please go ahead...

Georg 

-- 
!!!         Note, new phone numbers. Update your archives please
!!!
Georg Stellner
stellner@informatik.tu-muenchen.de+------------
Institut fuer Informatik, SAB    phone:  +49-89-289-22689          |
(())_-_(())
Technische Universitaet Muenchen fax:  +49-89-289-28232            |  |
(* *) |
D-80290 Muenchen                 room:   S 1211                    | {
\_@_/  }
http://wwwbode.informatik.tu-muenchen.de/~stellner                 |
`-----'

