I have a case that involves pulling a meshed rod through a bed of material that is loaded from an input file, not a restart file. When I change the number of processor divisions used the results of the force on the rod at the final state are significantly different. I have tried this with V3.10, 3.21 and 3.30, all have similar results. The only change to the input script was the number of processors alotted.
1 processor final force was -29.98 N
2 processors final force was -23.14 N
3 processors final force was -19.04 N
The pressure, stress and wear distributions on the rods are also significantly different.
Has anyone seen this behavior before?
ckloss | Thu, 09/24/2015 - 09:32
Hi,
Hi,
the final force will generally oscillate heavily - did you apply averaging? You should average at least for 1000 time-steps so that the results are comparable. Runs on different # procs should lead to statistically identical results
Christoph
jlstrom | Thu, 09/24/2015 - 16:26
Averaging
The final reported forces are averaged using fix all ave/time 10 80 800 for the 3 vector forces on the cylinder.
ckloss | Sun, 10/11/2015 - 16:58
Hi,
Hi,
if you post a simple test case I can have a short look
Christoph