Performance differences between RK2.5 and ARK2.5 in the latest version
TODO
-
RK2.5/ARK2.5 scalability benchmarking (with changing number of nodes, e.g. 10-24); check how MPI behaves
Intro
I compared the DCMPI41_mpi_40
experiment (generated by test/bash/create_runs.sh
) with only the time_scheme
parameter changed (RK2.5 or ARK2.5).
Results:
Average (after 5 runs):
- RK2.5: ~85 seconds
- ARK2.5: ~100 seconds
RK2.5
[ump84rg@jean-zay-pp1: DCMIP41_mpi_40]$ tail rk25/run_*/logs/gcm.log
==> rk25/run_0/logs/gcm.log <==
It No : 5399 t : 2591520.0
It No : 5400 t : 2592000.0
Time spent (s): 86 -- ms/step : 16.02 -- Throughput : 29966
Whole job (min) : 1 -- Completion in (min) : 0
GETIN restart_file_name = restart
masse advec mass rmsdpdt energie enstrophie entropie rmsv mt.ang
GLOB -0.111E-15 0.666E-15 0.17560E+01 -0.315E-04 0.531E-01 0.706E-05 0.690E+01 0.309E-04
Time elapsed : 86.7280650000000
Wed Apr 12 17:23:07 CEST 2023
==> rk25/run_1/logs/gcm.log <==
It No : 5399 t : 2591520.0
It No : 5400 t : 2592000.0
Time spent (s): 85 -- ms/step : 15.76 -- Throughput : 30461
Whole job (min) : 1 -- Completion in (min) : 0
GETIN restart_file_name = restart
masse advec mass rmsdpdt energie enstrophie entropie rmsv mt.ang
GLOB -0.111E-15 0.666E-15 0.17560E+01 -0.315E-04 0.531E-01 0.706E-05 0.690E+01 0.309E-04
Time elapsed : 85.2978970000000
Wed Apr 12 17:23:06 CEST 2023
==> rk25/run_2/logs/gcm.log <==
It No : 5399 t : 2591520.0
It No : 5400 t : 2592000.0
Time spent (s): 84 -- ms/step : 15.66 -- Throughput : 30648
Whole job (min) : 1 -- Completion in (min) : 0
GETIN restart_file_name = restart
masse advec mass rmsdpdt energie enstrophie entropie rmsv mt.ang
GLOB -0.111E-15 0.666E-15 0.17560E+01 -0.315E-04 0.531E-01 0.706E-05 0.690E+01 0.309E-04
Time elapsed : 84.7864500000000
Wed Apr 12 17:23:06 CEST 2023
==> rk25/run_3/logs/gcm.log <==
It No : 5399 t : 2591520.0
It No : 5400 t : 2592000.0
Time spent (s): 84 -- ms/step : 15.66 -- Throughput : 30660
Whole job (min) : 1 -- Completion in (min) : 0
GETIN restart_file_name = restart
masse advec mass rmsdpdt energie enstrophie entropie rmsv mt.ang
GLOB -0.111E-15 0.666E-15 0.17560E+01 -0.315E-04 0.531E-01 0.706E-05 0.690E+01 0.309E-04
Time elapsed : 84.7517700000000
Wed Apr 12 17:23:05 CEST 2023
==> rk25/run_4/logs/gcm.log <==
It No : 5399 t : 2591520.0
It No : 5400 t : 2592000.0
Time spent (s): 84 -- ms/step : 15.64 -- Throughput : 30692
Whole job (min) : 1 -- Completion in (min) : 0
GETIN restart_file_name = restart
masse advec mass rmsdpdt energie enstrophie entropie rmsv mt.ang
GLOB -0.111E-15 0.666E-15 0.17560E+01 -0.315E-04 0.531E-01 0.706E-05 0.690E+01 0.309E-04
Time elapsed : 84.6594900000000
Wed Apr 12 17:23:05 CEST 2023
ARK2.5
[ump84rg@jean-zay-pp1: DCMIP41_mpi_40]$ tail ark25/run_*/logs/gcm.log
==> ark25/run_0/logs/gcm.log <==
It No : 5399 t : 2591520.0
It No : 5400 t : 2592000.0
Time spent (s): 100 -- ms/step : 18.56 -- Throughput : 25866
Whole job (min) : 1 -- Completion in (min) : 0
GETIN restart_file_name = restart
masse advec mass rmsdpdt energie enstrophie entropie rmsv mt.ang
GLOB -0.111E-15 0.444E-15 0.17560E+01 -0.315E-04 0.531E-01 0.706E-05 0.690E+01 0.309E-04
Time elapsed : 100.412414000000
Wed Apr 12 17:27:49 CEST 2023
==> ark25/run_1/logs/gcm.log <==
It No : 5399 t : 2591520.0
It No : 5400 t : 2592000.0
Time spent (s): 100 -- ms/step : 18.57 -- Throughput : 25850
Whole job (min) : 1 -- Completion in (min) : 0
GETIN restart_file_name = restart
masse advec mass rmsdpdt energie enstrophie entropie rmsv mt.ang
GLOB -0.111E-15 0.444E-15 0.17560E+01 -0.315E-04 0.531E-01 0.706E-05 0.690E+01 0.309E-04
Time elapsed : 100.483556000000
Wed Apr 12 17:27:49 CEST 2023
==> ark25/run_2/logs/gcm.log <==
It No : 5399 t : 2591520.0
It No : 5400 t : 2592000.0
Time spent (s): 99 -- ms/step : 18.51 -- Throughput : 25933
Whole job (min) : 1 -- Completion in (min) : 0
GETIN restart_file_name = restart
masse advec mass rmsdpdt energie enstrophie entropie rmsv mt.ang
GLOB -0.111E-15 0.444E-15 0.17560E+01 -0.315E-04 0.531E-01 0.706E-05 0.690E+01 0.309E-04
Time elapsed : 100.158678000000
Wed Apr 12 17:27:47 CEST 2023
==> ark25/run_3/logs/gcm.log <==
It No : 5399 t : 2591520.0
It No : 5400 t : 2592000.0
Time spent (s): 100 -- ms/step : 18.57 -- Throughput : 25844
Whole job (min) : 1 -- Completion in (min) : 0
GETIN restart_file_name = restart
masse advec mass rmsdpdt energie enstrophie entropie rmsv mt.ang
GLOB -0.111E-15 0.444E-15 0.17560E+01 -0.315E-04 0.531E-01 0.706E-05 0.690E+01 0.309E-04
Time elapsed : 100.499062000000
Wed Apr 12 17:27:49 CEST 2023
==> ark25/run_4/logs/gcm.log <==
It No : 5399 t : 2591520.0
It No : 5400 t : 2592000.0
Time spent (s): 100 -- ms/step : 18.55 -- Throughput : 25869
Whole job (min) : 1 -- Completion in (min) : 0
GETIN restart_file_name = restart
masse advec mass rmsdpdt energie enstrophie entropie rmsv mt.ang
GLOB -0.111E-15 0.444E-15 0.17560E+01 -0.315E-04 0.531E-01 0.706E-05 0.690E+01 0.309E-04
Time elapsed : 100.407780000000
Wed Apr 12 17:27:49 CEST 2023
Follow up
@thomas.dubos Do we perform more test with specific experiments?
Edited by Patryk Kiepas