Convergent Science

Parallel Computing

The industry trend is clear: CFD will continue to require larger detailed chemical mechanisms with LES. Moreover, CONVERGE™ users continue to use large cell counts to promote grid convergent solutions. To accommodate this trend, Convergent Science, Inc. devotes extensive resources so that CONVERGE™ has cutting edge HPC capabilities.  CONVERGE™ CFD software is commonly run in parallel, both in shared and distributed memory architectures. 




When the SAGE detailed chemistry solver is used for combustion modeling (which is always recommended), CONVERGE™ parallelizes the flow independently from the chemistry. This allows the parallel performance to be optimized for both the combustion and CFD flow solver.

The initial cell partitioning is done automatically by CONVERGE™ at run-time. However, the initial partitioning will not be ideal as the simulation proceeds due to various factors including moving boundaries, adaptive mesh refinement (AMR), spray particles and ignition. CONVERGE™ can re-partition the domain automatically at user prescribed intervals.

To further improve parallel performance, the METIS load balancing algorithm (developed at the University of Minnesota) was implemented in collaboration with Argonne National Laboratory. METIS is available in the newest version of CONVERGE™ (v2.1). 




With the above features, the overall parallel performance of CONVERGE™ is excellent. That said, it should be pointed out that larger cell counts promote better parallel performance in general. 

The SAGE detailed chemistry solver scales almost perfectly in parallel. The larger the percentage of the overall solution is related to SAGE, the better the parallel performance. Therefore, using larger chemical mechanisms promote better parallel performance.





Share This Page