Dear Everyone,
I am running an FHIST compset in CESM2.2.0 with 0.9X1.25 resolution (288 longitudes and 192 latitudes) in a cluster where each node has 40 processors. The specification that I am using is MAX_TASKS_PER_NODE=40 and MAX_MPITASKS_PER_NODE=40. I have performed 2 runs of 1 year each, one with NTASKS=40 i.e. 1 node and another with NTASKS=160 i.e. 4 nodes. The first run took 36 hours and the 2nd one is still running and has taken 26 hours already. I can see that all 4 nodes are getting utilized then what is the probable reason of no reduction in the runtime to 9 hours. Do the number of processors need to be set as multiples of longitude?
Thank you in advance,
Shreya.
I am running an FHIST compset in CESM2.2.0 with 0.9X1.25 resolution (288 longitudes and 192 latitudes) in a cluster where each node has 40 processors. The specification that I am using is MAX_TASKS_PER_NODE=40 and MAX_MPITASKS_PER_NODE=40. I have performed 2 runs of 1 year each, one with NTASKS=40 i.e. 1 node and another with NTASKS=160 i.e. 4 nodes. The first run took 36 hours and the 2nd one is still running and has taken 26 hours already. I can see that all 4 nodes are getting utilized then what is the probable reason of no reduction in the runtime to 9 hours. Do the number of processors need to be set as multiples of longitude?
Thank you in advance,
Shreya.