In contrast to distributed computing, Message Passing Interface (MPI) computing requires a specialized high-performance computer cluster with a high-speed, low-latency interconnect such as InfiniBand®. However, the speed-up benefits provided by MPI computing can be immense, and it is the only way to simulate extremely large models that exceed the limits of a single machine.
With MPI computing, the simulation model is decomposed into multiple domains, and the workload is divided between nodes in the cluster. Because the calculations for each domain are not independent, the cluster computers need to transfer large amounts of data between simulation domains, hence the need for a dedicated high-performance interconnect. MPI computing is currently supported by the transient solver, integral equation solver and wakefield solver. With the transient solver, MPI computing can be combined with GPU computing for even greater performance benefits on compatible clusters.