Queues guillimin

Un article de Informaticiens département des sciences de la Terre et l'atmosphère
Version depuis le 14 d'avril 2014 à 21:05 par Katja (Discuter | changes)
Aller à: navigation, charcher

With the new phase2 of guillimin came a few new queues.

If you want to know which queues there are, their names, busyness, etc., execute the command:

    nodes

You will see there are essentially three types of queues:
    - 1 serial    Westmere queue for jobs with less than 12 cores
    - 3 parallel SandyBridge queues: 'debug', 'sw2', 'lm2'
    - 3 parallel Westmere queues: 'hb', 'lm', 'sw'

The names might look a little different in the 'nodes' output but you will know which one is which.

Good things to know about these queues:

    - Each Westmere node has two 6-core sockets => 12 cores per node. They are the old phase 1 nodes.
    - Each SandyBridge node has two 8-core sockets => 16 cores per node. They are the new phase 2 nodes.
    - Jobs using less than 12 cores will automatically go to the serial queue. And there is nothing you can do about it.
    - Jobs using OpenMP=3 or 6 get automatically send to a parallel Westmere queue ('hb', 'lm', 'sw')
    - Jobs using OpenMP=4 or 8 get automatically send to a SandyBridge queue ('sw2', 'lm2')
    - Jobs using OpenMP=1 or 2 get automatically send to a parallel Westmere queue OR a SandyBridge queue ('hb', 'lm', 'sw','sw2', 'lm2)
    - Jobs using OpenMP=5 or 7 will get refused.

If you want to send your job to a specific queue you can to so by setting the environment variable 'SOUMET_EXTRAS', for example:

    export SOUMET_EXTRAS="-q hb"

will send all jobs submitted (from the same window) using 12 or more cores to the 'hb' queue.
If you linked to the .group_profile you should already have a few aliases for this. You can check them with:

    alias | grep SOUMET_EXTRAS

BUT!!! Know that, if you send a job using OpenMP=3 or 6 to a SandyBridge queue or a job using OpenMP=4 or 8 to a Westmere queue, they will NEVER RUN!!!