Logo of Science Foundation Ireland  Logo of the Higher Education Authority, Ireland7 CapacitiesGPGPU Research Projects
Ireland's High-Performance Computing Centre | ICHEC
Home | News | Infrastructure | Outreach | Services | Research | Support | Education & Training | Consultancy | About Us | Login

User Mailing

ICHEC mail #7


Posted: 2005-10-20

Dear ICHEC users,


We would like to inform you of a number of changes in our scheduling policy, as well as provide some tips for a better utilisation of our resources.


Walton (IBM cluster 1350):
-------------------------

- CPUs from the development region have been re-assigned to the production region to increase our capacity for production runs. The total number of CPUs now allocated to the production region is 808 CPUs (340 nodes with 4Gb of RAM, and a further 64 nodes with 8Gb of RAM). This leaves approximately 120 CPUs for development jobs.

- the job size limit (maximum number of CPUs you may request for any single job) has been decreased to 256 CPUs to allow a better utilisation of our resources and faster turn-around time.

- the maximum runtime for the night queues (active daily from 6pm until 10am) has been raised to 15 hours (wall clock time).


Hints:
-----

1. A summary of our scheduling policies can be found at http://www.ichec.ie/scheduling_policy

2. The queue limits (maximum CPU count, maximum runtime, status, etc.) can be obtained with the command qstat -q, e.g.:

jcdesplat@l2cu27:~> qstat -q

server: torque.ichec.ie

Queue Memory CPU Time Walltime Node Run Que Lm State
---------------- ------ -------- -------- ---- --- --- -- -----
autoq -- -- -- -- 0 0 -- E R
ProdHiMemQ -- -- 120:00:0 8 1 5 -- E R
DevNightQ -- -- 15:00:00 16 0 1 -- E R
ProdLoMemQ -- -- 120:00:0 128 27 5 -- E R
DevDayQ -- -- 01:00:00 16 0 2 -- E R
superq -- -- -- -- 0 0 -- E R
--- ---
28 13


3. We recommend not systematically specifying the maximum allowable run-time in your PBS script as this is detrimental to effective scheduling. In particular, we encourage any user who may benefit from the night queue to do so (this queue is suitable for jobs requiring up to 32 CPUs, with a maximum runtime of 15 hours).

4. We would also ask everyone to be considerate for other users of the system, and in particular not to submit excessive numbers of jobs in the queueing system. The current policies are relatively permissive, and we would like to avoid having to enforce more restrictive rules.

5. Users currently specifying :ppn=1 in their PBS script (only using 1 CPU per node) are asked to avoid this practice. PBS has been configured to ignore their request and re-allocate their job so that it uses both CPUs on half the number of nodes specified in their script.


Finally, we would like to inform you that charging will be implemented on our systems from early November. We will send a user mailing a couple of days before this change takes place.

Return to User Mailings