----- "Sylvain Jeaugey" <sylvain.jeaugey_at_[hidden]> wrote:
> I must miss something. If the launcher did *not*
> do the binding, then their affinity mask is full, right?
Not necessarily, they could be on an Altix system which
has a boot CPU set of a 1 core that all the system daemons
tend to end up in, and everyone else ends up on the other
N-1 cores.
I think that the boot cpuset is on core 0 too, rather
than core N, so you need to work with cores 1->N rather
than 0->(N-1).
But from what Ralph has said on the OpenMPI list, it
sounds like they want this logic to be in OMPI rather
than in PLPA, so this might be academic..
cheers!
Chris
--
Christopher Samuel - (03) 9925 4751 - Systems Manager
The Victorian Partnership for Advanced Computing
P.O. Box 201, Carlton South, VIC 3053, Australia
VPAC is a not-for-profit Registered Research Agency
|
![[Flag]](../../../../../community/mirrors/flag-us.gif){kind=small}