MPI scalability (first results)

Parallelization with OpenMP and MPI, scalability, reproducibility, errors, problems suggestions
Post Reply
User avatar
Jaakko Leppänen
Site Admin
Posts: 2377
Joined: Thu Mar 18, 2010 10:43 pm
Security question 2: 0
Location: Espoo, Finland
Contact:

MPI scalability (first results)

Post by Jaakko Leppänen » Tue Dec 27, 2011 12:52 pm

Here are the first results for parallelization with MPI... The calculation cases are the same as in the previous OpenMPI scalability test. The MPI routines still require some fine-tuning and work on repruducibility, but this is what we get with similar non-reproducible approach as in Serpent 1:

Image
Figure 1. Speed-up factor as function of number of MPI tasks in LWR cases. Transport calculation only.

Image
Figure 2. Speed-up factor as function of number of MPI tasks in CANDU, HTGR and SFR cases. Transport calculation only.

The scalability seems to be comparable with that seen with OpenMP, but I expect the performance to improve in some burnup calculation problems, since the use of distributed memory parallelization probably improves the scalability of burnup and processing routines.

I'll post more results when I get the hybrid mode fully working...
- Jaakko

Post Reply