On Tue, 20 May 2003, Jerome BENOIT wrote:
> I noticed the fellowing behaviour when a LAM session is initialized with
> a given priority
> [ nice -n 15 lamboot <lamboot_args> ],
> the processes running on the original node inherit the priority, but the
> processes running on the other nodes get the normal priority. This
> behaviour sounds odd:
It's a direct result of the fact that rsh/ssh (and potentially other boot
mechanisms) do not propogate the priority to remote processes.
> it would be nice to allow `lamboot' to propagate through the cluster its
> priority. In such a way the user can easily set a priority to each
> LAM_SESSION.
That will definitely take some work to implement, because it will likely
somewhat change the architecture of how LAM boots itself (off the top of
my head: remote launch a process, send the process the desired priority,
have the process nice(2) itself). It may get more complicated because
some boot mechanisms may actually preserve the priority (e.g., bproc) and
not need these extra steps. This is not necessarily rocket science, but
it will take a bit of care to get it Right, and would likely need to be
incorporated into the boot SSI protocols for all modules.
This certainly won't happen for 7.0, but we can add it to the "to-do" list
for possible future versions.
As a workaround, you might want to simply add a command line parameter or
environment variable to your MPI executables and nice(2) that value.
--
{+} Jeff Squyres
{+} jsquyres_at_[hidden]
{+} http://www.lam-mpi.org/
|
