On 22/05/18 17:11, Xiaoye S. Li wrote: > Numerical factorization is always parallel (based on number of MPI > tasks and OMP_NUM_THREADS you set), the issue here is only related to > symbolic factorization (figuring out the nonzero pattern in the LU > factors). Default setting is to use sequential symbolic factorization, > precisely due to the ParMETIS bugs. > (The parallel symbolic factorization needs ParMETIS ordering.) > > Question for PETSc developers -- > Is PT-Scotch an installation option? I can try to use PT_Scotch > together with parallel symbolic factorization, which should be more > stable than ParMETIS.
--download-ptscotch is a thing (I can't remember what other bits it needs). Cheers, Lawrence
