[torqueusers] Post-job queueing process

Ronny T. Lampert telecaadmin at uni.de
Thu Sep 29 02:46:51 MDT 2005


> 1) What actually triggers the scheduler (assuming the default
The scheduler is triggered via the pbs_server, not by itself.
That also means you'll trigger it when you submit a new job.

The key you are maybe looking for is:

qmgr> set server scheduler_iteration = <SECONDS>

> 2) What is the alarm_time in the scheduler for?
How long a scheduling run may take.

If you have a _LARGE_ number of jobs you may want to increase this time,
else the _scheduler restarts (via exec()) itself.
In a scheduling run all jobs descriptions are fed to the scheduler.

I had problems, because I am installied PBS into a custom dir, it's path not
in the $PATH.

> 3) The find_best_node() routine tries to find nodes for a job only when
> the job itself does not request nodes resource. In this case when the
> scheduler does a pbs_runjob request to the server, it does not specify a
> node to run job on? So how would the scheduler know whether there are
> enough nodes available in the system?
I dont know enough to answer that.
I though the scheduler decides which job to run next and the server mapps
the job onto the physical nodes.


More information about the torqueusers mailing list