LAM: How to launch more processes?

Brian Barrett <brbarretlam-mpi.org> wrote:

On May 5, 2006, at 5:18 PM, imran shaik wrote:

&gt; I am trying to launch as many MPI processes as possible to collect
> few performance statistics.
>
> How to launch many processes using lamboot.?
>
> I need around 250.
>
> I have three P4 machines with 256MB ram connected via hub.

With LAM/MPI, you can't do it. LAM/MPI was designed for HPC
applications running on small machines. HPC apps rarely run
oversubscribed, and only then by a couple nodes. LAM/MPI can start
about 60-70 processes / node, so that would give you 180 - 210
processes on your three machines. Of course, I can't see how you're
going to start that many processes on that small of a configuration,
but that's another story.


Brian

--
Brian Barrett
LAM/MPI developer and all around nice guy
Have a LAM/MPI day: http://www.lam-mpi.org/


_______________________________________________
This list is archived at http://www.lam-mpi.org/MailArchives/lam/

Imran

Brian Barrett <brbarretlam-mpi.org> wrote:

On May 13, 2006, at 10:36 AM, imran shaik wrote:

&gt; Can you suggest me any way of solving my problem?
&gt; Shall i move away from MPI?
>
> Any suggestions are welcome!
&gt;
>
&gt; Awaititng for your reply.
>;
> PS:
> I am doing an academic project on dcentralized grid resource
> discovery. I need to evaluate(simulate) a request forwarding
> algorithm.I am planning for as many processes as i can with the
> available resources. I chose MPI coz it provided with facilities
> such as communicators and topology . Now i am learning MPI and
> almost about to complete.I heard of people doing simulations using
> MPI processes, perhaps with more nodes. can u suggest any alternative?
>
>; Any suggestions ,positive or negative are most welcome.

It might be possible to use another MPI implementation (Open MPI or
MPICH) and get the process count you're looking for. Of course, I
think you're going to find that your performance is useless -- An MPI
implementation needs to keep a bunch of data around and you don't
have much memory. On the other hand, it might do what you need (swap
can do an awful lot if needed). I can't answer for you if MPI is the
right thing or not. In general, MPI isn't good at decentralized,
dynamic environments. But if it's doing what you need, then why redo
the work you've already put into your prototype?

Brian

--
Brian Barrett
LAM/MPI developer and all around nice guy
Have a LAM/MPI day: http://www.lam-mpi.org/


_______________________________________________
This list is archived at http://www.lam-mpi.org/MailArchives/lam/

__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com