All that LAM will do is launch 2 copies of foocode on your machine.
LAM doesn't have any processor-binding code. So if your 2 copies of
foocode are staying on one core, there's something else going on.
On Jul 1, 2009, at 1:35 PM, Rahul Nabar wrote:
> While trying to run LAM jobs on my dual core laptop I was facing
> problems and was looking for some debugging sugesstions.
>
> Symptoms:
>
> Single cpu jobs run fine (even give me the correct answers with some
> test scripts) BUT dual cpu jobs keep running on single cpus. Same if I
> try higher "oversubscribed" runs.
>
> recon and lamboot give no errors.
> lamnodes shows a single node but with 2 cpus.
>
> I start with:
>
> lamboot -d -v bhosts
>
> bhosts:
> x.x.x.y
> x.x.x.y
>
> (where x.x.x.y is my IP address)
>
> What could be going wrong if lamnodes shows 2 cpus but my mpicode
> still runs on one
>
>
> THe code is in C and compiled with mpicc. I invoke it with:
>
> mpirun -np 2 foocode
>
> Any suggestions what I should be looking for? [Also, just to make sure
> my C code was not buggy I ran it on another well-tested server and
> there it ran fine for all numbers of CPUs. So the problem seems to be
> in the way I have set up this particular architecture]
>
> Of course, if it helps I can post my C code or any other files too.
>
> --
> Rahul
> _______________________________________________
> This list is archived at http://www.lam-mpi.org/MailArchives/lam/
>
--
Jeff Squyres
Cisco Systems
|
