Flux update during an MPI run between decomposed case parts?
I am running a large-ish case as a one off and it is quite large so ends up running back and forward to Swap Drive and hence is taking forever.
The results are looking quite patchy (so far) and it looks as though the information from one processor domain is not being passed through to the next processor domain very well.
The case is merely 100 iterations young, but I would still expect less of a variance when looking at results than what I am getting.
My question is, how often are the fluxes transferred/updated between the processor domains?
A second question if I may, would it be possible to run each decomposed processor case as a seperate simulation (simpleFoam) and then once I have done a heap of iterations and am happy with each processor, then run it as one single case using MPI run to smooth out the result?
I would only be doing this as I know running each case would be a lot faster in serial, than running the whole lot in parallel at once. I have a 16 million cell case and merely 8GB ram so it is going to my 10k Raptor HDD as swap.
|All times are GMT -4. The time now is 23:20.|