Opened 11 years ago

Closed 10 years ago

#438 closed help (fixed)

Optimal settings for runs on Hector

Reported by: aschurer Owned by: lois
Component: HECToR Keywords:
Cc: Platform:
UM Version: 4.5

Description

Hi, I am about to begin several long runs on Hector using the UM vn4.5, totaling several thousands of years.
Consequently it is very important that these runs are completed as quickly and efficiently as possible. I would therefore be grateful for any advice about the optimum number of CPUs to use East-West and North-South.
In addition I want to maximise the time that simulations are running for compared to the amount of time spent queuing, so any recommendations as to the best target run length to use would be very useful.
Thanks,
Andrew

Change History (2)

comment:1 Changed 10 years ago by lois

  • Owner changed from um_support to lois
  • Status changed from new to assigned

Sorry Andrew, your query seems to have fallen into a 'black hole'.

You don't say what experiemnt you are running using vn4.5 but if it is more or less standard HADCM3 then the standard HECToR job (on the PUMA UMUI under the userid umui) uses N-S 4 and E-W 8 so 32 cores in total. On the XT4 changing the ratio of N-S to E-W for such a small job makes not really much difference in timings.

You also ask how to get through the queues quickly, this is a challenge on a shared machine! The XT4 is very very full at the moment as they have halved the overall size of the machine in order to install the XT6 - hence the long queues. The XT6 is now installed which could reliev pressure but they are being run as separate systems until e-lustre is installed, some time in June hopefully. As there are short queues -20 mins and 1 hour - and long queues 3, 6, 12 hours it really is a question of judgement as to whether it is better to continuosly resubmit (using automatic re-submission of course) to short queues or reduce re-submission as much as possible to longer queues. It is hard to give advice, it will have to be up to your judgement.

Sincere apologies for the delay in replying to your query.

Lois

comment:2 Changed 10 years ago by lois

  • Resolution set to fixed
  • Status changed from assigned to closed
Note: See TracTickets for help on using tickets.