Newsgroups: comp.parallel.mpi
From: name@domain.com (Cy Berman)
Subject: Re: MPICH with R4400 (SMP) and R8000
Organization: DisOrganized
Date: Fri, 20 Sep 96 15:25:44 GMT
Message-ID: <51ud55$556@news.erinet.com>

In article <y21afulgx79.fsf@como.arc.umn.edu>, Barry Bolding <barry@como.arc.umn.edu> wrote:
>
>We are setting up mpich 1.0.12 to run on a cluster of workstations
>with the primary machine being a multiprocessor ONYX and most of the
>other machines being either R4000 or R8000 SGI's. We are having some
>problem getting cross platform applications to run. To summarize, all
>libraries will be located on a nfs mounted file system
>
>/usr/local/mpi
>
>Beneath this will be the different library subdirectories for the different 
>platforms.
>----------------
>I build for single processor SGI's with the following
>
>configure -device=ch_p4 -arch=IRIX -mpe
>
>and all the R4000 SGI's(except the multiprocessor ONYX) can start jobs 
>and distribute them to other machines (including the ONYX)
>On the ONYX I get a 
>PERMISSION DENIED
>error when I execute the mpirun, and it seems to hang until I CNTRL-C, 
>which gives some SIGINTS
>
>> mpirun -np 2 a.out
>Permission denied.
>bm_list_13343:  p4_error: interrupt SIGINT: 2
>p0_13342:  p4_error: interrupt SIGINT: 2
>
>
>
>----------------
>I can build on the ONYX with the following configuration
>
>configure -device=ch_shmem -arch=IRIX -mpe
>
>and the ONYX runs jobs fine but cannot communicate correctly with other 
>machines.
>
>----------------
>If I configure on the ONYX with the following
>
>configure -device=ch_p4 -arch=IRIX -comm=shared -mpe
>
>The installation manual states that this should be able to function 
>in a workstation cluster or in smp mode, but I get similar behavior 
>to that for a straight ch_p4 build without the comm=shared, in 
>other words I get the PERMISSION DENIED error when running on the ONYX.
>
>Any ideas or suggestions would be appreciated.
>
>Barry
>

