To run MPI NUBEAM, in your Namelist simply set:
nbi_pserve = 1 ! for MPI run with NUBEAMIf you use "tr_start" or "xtranspin" the script will detect this setting and ask you how many processors you want.
It should be noted that MPI imparts no benefit if the problem size is too small. As a rough guide, we recommend 4000 particles per processor. I.e. for NPTCLS=1000 to 5000 use a standard serial TRANSP run.
For NPTCLS=8000 -- Np = 2 For NPTCLS=16000 -- Np = 4 For NPTCLS=32000 -- Np = 8
The more processors requested, the longer the queue wait-- users may find that Np = 16 or Np = 8 will be a practical upper limit on the existing PPPL computational cluster.
For some runs, MPI may help with smaller particle counts. This will certainly be the case if the low field enhanced FLR model is activated (i.e. if NLBGFLR=.TRUE. as is used in some ST simulations).