No speed-up with hycm_std
Posted: February 8th, 2022, 11:46 am
Hello forum,
Before posting, I looked over existing threads on this topic, but none have solved my problem yet.
I am running dust dispersion runs using ichem=3 on our linux cluster with 28-core nodes, using the pre-compiled HYSPLIT installation hysplit.v5.1.0_RHEL6.10. When I try running hycm_std (parallel) in place of hycs_std (serial), I see no speed-up or slight degradation in performance, no matter how many processors I specify (2, 4, 8, etc.). I tried varying mgmin between 1 and 300 (according to another thread), but this doesn't impact performance either. I'm running as "mpirun -np # ../exec/hycsm_std", where # is the number of processors. FYI, I have confirmed that I see multiple jobs running when doing a "top" command.
Any assistance is greatly appreciated! Do I need to consider obtaining the source and compiling on our system to realize the parallel performance improvement? I have included the SETUP.CFG, MESSAGE, and CONTROL files.
Sincerely,
Jonathan Case
NASA/SPoRT Center
Before posting, I looked over existing threads on this topic, but none have solved my problem yet.
I am running dust dispersion runs using ichem=3 on our linux cluster with 28-core nodes, using the pre-compiled HYSPLIT installation hysplit.v5.1.0_RHEL6.10. When I try running hycm_std (parallel) in place of hycs_std (serial), I see no speed-up or slight degradation in performance, no matter how many processors I specify (2, 4, 8, etc.). I tried varying mgmin between 1 and 300 (according to another thread), but this doesn't impact performance either. I'm running as "mpirun -np # ../exec/hycsm_std", where # is the number of processors. FYI, I have confirmed that I see multiple jobs running when doing a "top" command.
Any assistance is greatly appreciated! Do I need to consider obtaining the source and compiling on our system to realize the parallel performance improvement? I have included the SETUP.CFG, MESSAGE, and CONTROL files.
Sincerely,
Jonathan Case
NASA/SPoRT Center