I have also been unable to compile MPICH successfully with -with-comm=shared
configuration option.
If you compile your F and F90 files using the threadsafe versions (xlf_r and
xlf90_r) and including your MPICH libraries, you may be able to fire off a
shared-memory version (I've seen up to 198% cpu utilization per node) by
just assigning one job per node. If your F and F90 files were compiled with
xlf and xlf90, only one cpu will be used and you'd have to initiate two MPI
jobs per node.
Jan
Dr. Johannes N. Theron
Theron Engineering Solutions
4316 Rebel Circle
Cottonwood, AZ 86326
Phone: (928) 274-2905
email@hidden
Adjunct Professor
Department of Mechanical Engineering
College of Engineering and Technology
Northern Arizona University
P.O. Box 15600
Flagstaff, AZ 86011
email@hidden
(928) 523-6985
Don't forget that the machine files are a little bit different for
MPICH. If you've followed my postings on the scitech list you'll know
that I've asked for assistance on the machinefiles. Anyway, long story
short, for MPICH to run 1 task on each processor you need to list each
Xserve hostname twice and the machine from which you launch the job
only once in the machinefile.
Good Luck!
-Warner
Warner Yuen
Research Computing Consultant
Apple Computer
email: email@hidden
Tel: 408.718.2859
Fax: 408.718.0133
On Sep 18, 2004, at 12:04 PM, email@hidden wrote:
> Message: 2
> Date: Sat, 18 Sep 2004 10:20:29 -0500
> From: "Chona S. Guiang" <email@hidden>
> Subject: problem with MPICH-P4
> To: <email@hidden>
> Message-ID:
>
> <email@hidden>
>
> Content-Type: text/plain; charset="iso-8859-1"
>
> Hi,
>
> I built the latest version of MPICH on our G5 cluster. At runtime,
> I get the following error:
>
> stampede 34> bsub -I -n 4 ./mpichp4_wrapper ./mpimd
> Job <2302> is submitted to default queue <normal>.
> <<Waiting for dispatch ...>>
> <<Starting on compute-1-9>>
> Reason : Cannot allocate memory
> p0_7474: p4_error: OOPS: shmat failed : 1
>
> FYI, SHMMAX is set to 32MB in /etc/rc, and I've tried higher
> values as well, but to no avail. I've also used cleanipcs to get rid
> of shared memory segments and semaphores that were in use. The
> configure options for MPICH were:
>
> --with-device=ch_p4 \
> -comm=shared \
> -cc=xlc \
> -c++=xlc++ \
> -f90=xlf90 \
> -f77=xlf \
> -rsh=ssh
>
> Please let me know if you have any ideas on how to fix this. I
> looked at the MPICH FAQ and couldn't find anything. A google search
> was not very productive either.
>
> Thanks in advance.
>
> Chona
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: text/enriched
Size: 2218 bytes
Desc: not available
Url :
http://lists.apple.com/mailman/private/scitech/attachments/20040918/15745ade/attachment.bin
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Scitech mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
http://lists.apple.com/mailman/options/scitech/email@hidden
