Fluka parallel run without Flair

Dear Fluka experts,

Is it possible to run Fluka without Flair in multi core mode? In my project there is about 1000 input files, because of it Flair is not an option since it will take a lot of time to run them manually. Is there an option to run Fluka in MCNP style (one input file runs parallel on multiple cores by using MPI), or Flair style (multiple clones of input file runs independently on separate cores) is the only option?

Thank you in advance

Dear @Alexy

According to this older post http://www.fluka.org/web_archive/earchive/new-fluka-discuss/2521.html it seems that spawning yous jobs like Flair does is the only option. However, you could do this with a simple bash script that creates new input files with different random number seeds so you don have to open Flair for every job.


Dear @Alexy,
@lorenzo.mercolli is right, FLUKA doesn’t support MPI, so the only option is to make 1000 clones with different random number and submit them.
There are several possibilities through flair:

  1. Create manually a script that alters the RANDOMIZe card and creates a new input file
  2. Use flair to create the inputs (without running them).
  • Configure in the preferences the filename pattern of the jobs (it can be even on different directories look the flair manual)
  • Set as spawn the number of jobs
  • Select as queue the null instead of the Default
    This way it will create only the input files with different random number
  • Click on Run, to ONLY create the files and not run them.
    (Maybe the null I should rename it to Save :slight_smile:
  1. You can create your own queues and add them in the ${HOME}/.flair/flair.ini, e.g. a submit.sh script that sends the jobs to a cluster or other nodes.
  2. You can also use the tsp (task spooler) as queue. This is a primitive batch system, that allows to run N=number of cores concurrent jobs on the same machine. So if you submit 1000 jobs on a 32core machine it will run them in batches of 32 jobs at a time.

The advantage submitting through flair is that it will generate all the merging rules for you.
Even if you don’t run through flair, if you set the appropriate filename pattern, and the # of spawn jobs flair will create the rules for you anyhow.

Thank you for the tips. I’ll try to make Python multiprocessing file and run several input files in parallel.