On Apr 19, 2008, at 11:06 AM, James Forester wrote:
> I am running lam on a Debian Linux Beowulf cluster (ultimately
> running Rmpi in the R statistical environment). The problem I am
> running into is that the "nice level" of the spawned nodes is always
> set to 0. I found only one mention of this problem in the archives 7
> years ago (saying that it was nearly fixed): http://www.lam-mpi.org/MailArchives/lam/2001/12/3656.php
> .
>
> Is there now a way to specify nice levels on the slave nodes?
mpiexec was never ported to LAM/MPI and we never worked around this
"feature" of PBS inside of LAM/MPI. LAM actually couldn't do anything
more about the situation than you could by calling nice() sometime
early in your application, as it's entirely user level. This is
really something that the patch scheduler should be worrying about, as
LAM just inherits the nice level provided by the scheduler's daemon.
Brian
--
Brian Barrett
LAM/MPI Developer
Make today a LAM/MPI day!
|
