Load imbalance in density/force routines (after splitting?) #788
Description
In some of @msha0023 's calculations we are getting cpu/wall of only 9.4 when running on 64 cpus, indicating severe load imbalance. This may be related to use of particle splitting?
input file split4.in written successfully.
Since code start: 65102 timesteps, wall: 4.55E+05s cpu: 1.29E+07s cpu/wall: 28
Since last dump : 9532 timesteps, wall: 6.40E+04s cpu: 1.73E+06s cpu/wall: 27
| wall | срu | cpu/wall | load bal frac | ||
|---|---|---|---|---|---|
| step | 17.78h | 480.87h | 27.04 | 100.00% 99.95% | |
| -tree | 5.85h | 210.07h | 35.88 | 100.00% 32.91% | |
| -density | 5.19h | 48.94h | 9.43 | 100.00% 29.16% | |
| -force | 4.28h | 84.08h | 19.63 | 100.00% 24.07% | |
| cons2prim | 0.14h | 8.52h | 62.41 | 100.00% 0.77% | |
| Substep | 1.50h | 88.93h | 59.27 | 100.00% 8.43% | |
| -sink-sink | 0.00h | 0.00h | 37.33 | 100.00% 0.00% | |
| -gas_force | 0.22h | 12.96h | 58.79 | 100.00% 1.24% | |
| -accretion | 0.71h | 40.97h | 57.50 | 100.00% 4.01% | |
| write_ev | 0.01h | 0.46h | 58.45 | 100.00% 0.04% | |
| write_dump | 0.00h | 0.00h | 16.00 | 100.00% 0.00% |