I'm running a simulation of a meteoritic impact with roughly 1.5M particles using mpi with 16 cores. It shows a reproducible instability at the boundaries between sub-domains. See below:
Pressure:
MPI boundaries:
This effect doesn't show up without parallelization. And persists changing initial conditions (in different parts of the domain, but always at subdomains' boundaries).
I tried changing the number of ghost layers and the parallel scale factor. I checked my equations' groups both with real = False and real = True, and also did the same with the update_nnps parameter. I also tried changing resolution and kernel function.
I was wondering how to remove or mitigate the issue.
Thank you in advance.
Marco
I'm running a simulation of a meteoritic impact with roughly 1.5M particles using mpi with 16 cores. It shows a reproducible instability at the boundaries between sub-domains. See below:
Pressure:
MPI boundaries:
This effect doesn't show up without parallelization. And persists changing initial conditions (in different parts of the domain, but always at subdomains' boundaries).
I tried changing the number of ghost layers and the parallel scale factor. I checked my equations' groups both with real = False and real = True, and also did the same with the update_nnps parameter. I also tried changing resolution and kernel function.
I was wondering how to remove or mitigate the issue.
Thank you in advance.
Marco