[torquedev] Ommiting core 0 from /dev/cpuset/torque/cpus breaks cpusets

Chris Samuel csamuel at vpac.org
Sun Apr 20 07:51:23 MDT 2008


----- "Chris Samuel" <csamuel at vpac.org> wrote:

> Hi folks,
> 
> It looks like if you pre-create /dev/cpuset/torque with
> cores 1-n to cope with your system allocating core 0 for
> a boot cpuset the current code doesn't cope and the jobs
> end up with no cpus allocated to the job cpuset.
> 
> I've verified this with my current build but not had a
> chance to check with trunk yet.

OK - confirmed against trunk.

My guess is that it is because currently we assume that
there is a one to one mapping from vnode to core, which
isn't appropriate if someone has reserved anything other
than the final core (and only then if pbs_server thinks
that the box has n-1 cores).

The solution is probably to parse the information in
/dev/cpuset/torque/cpus and have an array of vnodes->cores
and use that instead.

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


More information about the torquedev mailing list