Newsgroups: comp.parallel.mpi
From: gdburns@osc.edu (Greg Burns)
Subject: Re: Preliminary Program: MPI Developers Conference
Organization: Ohio Supercomputer Center
Date: 13 Jun 1995 14:06:12 -0400
Message-ID: <3rkk2k$5ur@tbag.osc.edu>

In article <3rk6mt$1sf7@info4.rus.uni-stuttgart.de> zrf90396@awsws8.rus.uni-stuttgart.de (J. Scheurich) writes:
>Hi,
>
>We are thinking about using MPI as a fast 8-) and save 8-( portable 
>peer-to-perr -packet in heterogeneous network.
>Therefore we are miss some features inside the current level of MPI 
>(e.g. what is called in pvm a way to detect a splitting of the virtual 
> machine cause of network problems)
>It's difficult to see for us, if the next level of the MPI-standart 
>(MPI-2) will include such features.
>There are some interesting points on the conference:
>
>>     [*] Greg Burns, Raja Daoud
>>         Robust MPI Message Delivery through Guaranteed Resources

This paper is not really about the fault tolerant features you describe.
It's about delivering messages in compliance with MPI and how portable
applications can assess the inherent delivery limitations of a particular
implementation.

It's our other talk that will touch upon fault tolerance because this
is introduced in LAM 6.0.  Nodes can die and applications can detect
this.  Rippling this up to MPI applications is not so easy, but we
do present a basic method for writing fault tolerant MPI master/slave
applications in conjunction with an early implementation of MPI_Comm_spawn().

>Are there some documents available about this points ?
>Will there are some documents after the conference ?

I understand all papers will be due at the conference.

-=-
Greg Burns				gdburns@tbag.osc.edu
Ohio Supercomputer Center		http://www.osc.edu/lam.html
MPI Developer's Conference, June 21-23 @ Notre Dame

