On Wed, Oct 27, 2004 at 09:26:28AM +0100 or thereabouts, Stephen Childs wrote:
> I am trying to configure PBS so that each queue is associated with a subset
> of nodes. The setup is as follows:
>
> Node1 doesn't support dteam users, Node2 does.
>
> In my node config I have:
>
> node1 np=2
> node2 np=1 dteam
The reason this does not work is that the job manager
/opt/globus/setup/globus/lcgpbs.in
/opt/globus/lib/perl/Globus/GRAM/JobManager/lcgpbs.pm
contains lines that sends things like
#PBS -l nodes=1\n
If you want to add different job requirements for different queues the only
option today is the hack the lcgpbs.in file.
Steve
>
>
> and in my PBS config I have:
>
> Queue: dteam
> queue_type = Execution
> total_jobs = 0
> state_count = Transit:0 Queued:0 Held:0 Waiting:0 Running:0 Exiting:0
> resources_max.cput = 48:00:00
> resources_max.walltime = 72:00:00
> resources_default.neednodes = dteam
> resources_default.nodect = 1
> resources_default.nodes = 1
> acl_groups = dteam
> enabled = True
> started = True
>
> If I submit manually using either qsub -l nodes=1:dteam or qsub -q dteam
> then the job goes correctly to node2. If I submit via globus using
> "globus-job-run gridgate.cs.tcd.ie/jobmanager-lcgpbs -q dteam" then I get no
> output (I think because it is being sent to node1 which can't run dteam
> jobs.) If I reverse the order in the nodes config file, then the job runs
> successfully on node2 -- PBS seems to be just choosing the first node
> regardless of its properties.
>
> Any ideas? (I know this has been asked on the list before but no solution
> was given.) It seems (from the earlier thread) that the problem is globus
> generating a submission akin to "qsub -q dteam -l nodes=1" and that the
> "nodes=1" is overriding any defaults set in the PBS configuration.
>
> Stephen
--
Steve Traylen
[log in to unmask]
http://www.gridpp.ac.uk/
|