Reproducibility in parallel mode

This category replaces the missing input manual
Post Reply
User avatar
Jaakko Leppänen
Site Admin
Posts: 2380
Joined: Thu Mar 18, 2010 10:43 pm
Security question 2: 0
Location: Espoo, Finland
Contact:

Reproducibility in parallel mode

Post by Jaakko Leppänen » Mon Feb 06, 2012 4:36 pm

Serpent 2 should be able to reproduce the transport simulation in serial and parallel modes using the same random number seed. The reproducibility option is set by:

Code: Select all

set repro <mode>
where <mode> has the following options:

0 - No reproducibility
1 - Reproducibility with OpenMP parallelization
2 - Reproducibility with MPI and hybrid OpenMP / MPI parallelization

Option 1 is the default, and it should work in both transport and burnup modes. Setting the option to 0 may speed-up some calculations where the population size is large.

Option 2 is under development, and it seems to work in most cases. It should be noted, however, that a parallel calculation run with option 2 reproduces a serial calculation run with option 2, not a serial calculation run with option 1 or 0. This calculation mode can be slow, and it is mainly intended for debugging purposes.

Also note that reproducibility requires running the calculation with the same random number seed, i.e. setting the seed manually:

Code: Select all

set seed <N>
or using the "-replay" -command line option after a previous run with the same input file.
- Jaakko

Post Reply