Hi Ross,
The code is dead simple so I do not think there is a bug. Also it seems that
others are having no problem using mpich and lam on machines with a single cpu
so I think you are right about the configuration being screwy. Could you be
more specific with the ssi options that may be useful? I would appreciate it.
Thanks... Rich
Quoting Ross Boylan <ross_at_[hidden]>:
> On Wed, 2006-02-15 at 14:18 +0100, rtichy_at_[hidden] wrote:
> > Hi,
> >
> > Quoting Esteban Fiallos <erf008_at_[hidden]>:
> >
> > > Well, it depends on what your program is supposed to do. If, for
> > > instance, you are performing a send and receive operation that
> > > requires more than one processor, then the program will crash.
> >
> > It does exactly this... rank 0 id the 'master', rank 1 is the 'foreman' and
> all
> > higher ranks are workers. Messages are sent to different agents depending
> on
> > their rank. Is there any way to force lam to asign different ranks on one
> > processor?
> >
> > -Rich
> My non-expert understanding is that each process has a different rank,
> regardless of what processor they are on.
>
> I don't think lam should crash just because everything is on one
> processor.
>
> The program could lock up if you've made some timing assumptions that
> are exposed by everything being on one processor.
>
> When communicating within the same box lam can use different methods
> (e.g., shared memory) than it would use between boxes. Possibly there
> is a bug in that code, it's misconfigured, or you need to set an option.
> I think some of the ssi options control this.
>
> Ross Boylan
>
> _______________________________________________
> This list is archived at http://www.lam-mpi.org/MailArchives/lam/
>
|
