Closed ziu1986 closed 4 years ago
Scaling test has been conducted Date: 28.01.2020 CPUs: 8, 16, 32 Model setups: Default, mOSaic, with standard output/additional output of non-transported tracers Tracers tot: 110 Model integration: 10 days
No. CPUs | Average | Std |
---|---|---|
8 | 1.41 | 0.09 |
16 | 0.45 | 0.02 |
32 | 0.147 | 0.015 |
After we have moved to SAGA, the old scaling tests done by Amund on ABEL may no longer hold. The reason for this is among other things a differing architecture: CPU/node ABEL SAGA 16 40 In the default configuration, running one model day takes roughly 60/12 min on 1/40 CPU(s). Asking for too many resources will affect the queueing time on SAGA negatively. We should do a scaling test: 5x5 model days for different numbers of CPUs (1,4,8,16,24,32,40)