MPI/OMP Hybrid Multi-threading

EFDC+ has been optimized to apply a domain decomposition approach based on Intel’s Message Passing Interface (MPI). With this feature, EFDC+ can run across large cluster systems as well as multi-core desktop computers with greater speed than ever before.  EFDC+ with OMP typically produced run times up to 4 times faster on a six-core processor than the conventional single-threaded EFDC model.  Testing on large domain models (500K cells) demonstrates that this new hybrid MPI/OMP approach achieves a speedup of nearly 17 with 32 compute cores. Using a cluster configured on Amazon Web Services (AWS), a strong scaling study was conducted, which resulted in a speedup of nearly 25 while using 96 compute cores.

Overall, the domain decomposition approach using MPI provides the ability to run across a cluster and offers substantially improved performance compared to previous versions of EFDC+. To learn about the performance gains achievable with this new approach, as well as guidance on configuring models read, our white paper

Overview of EFDC+’s hybrid MPI/OMP domain decomposition.
EFDC+ speedup and the fraction of total computational time spent in communication relative to the number of cores used.