Newsgroups: comp.parallel.mpi,comp.parallel.pvm
From: lusk@donner.mcs.anl.gov (Rusty Lusk)
Subject: Re: Problems by running mpirun
Organization: Argonne National Laboratory
Date: 2 Dec 1996 00:19:10 GMT
Message-ID: <57t79u$ru9@milo.mcs.anl.gov>

In article <329DA7EA.28D8@fz-rossendorf.de>, Andreas Haehne <haehne@fz-rossendorf.de> writes:
|> #######       CALL FOR HELP    #############
|> 
|> To run the program TEST on a cluster consisting of workstations from HP
|> and IBM I use
|> the skript mpirun as descripted in the MPICH User's Guide.
|> 
|> 	----> mpirun -arch hpux -np 2 -arch rs6000 -np 2 TEST
|> 
|> In this manner I'm able to start the program on workstations with
|> different architectures.
|> I do this by using the same device, inthis case ch_p4.
|> 
|> BUT I don't know, how I can run an application on systems with the same
|> architecture and 
|> different devices. Well, I'm trying to start a program on HP
|> workstations and on a Convex SPP1000 machine.
|> Both kinds of machines rely on the architecture hpux but use different
|> kinds
|> of devices. The SPP-machine uses ch_shmem and HP workstations use ch_p4.
|> 
|> 	
|> ---> How can I run a program in this case, without installing the
|> routines for the ch_p4
|>      device on the SPP?

You might try configuring on the SPP with

  -device=ch_p4 -comm=shared

Then you will get the multiprotocol version on the SPP that can send
messages internally on the SPP via shared memory, and externally via
TCP-IP.  You will want to use a procgroup file to specify the locations of
processes to mpirun.

