POLCOMS - parallel computing

The present state-of-the-art computing is parallel, that is, problems are solved on clusters of processors linked by high speed communication links. Efficient communications between processors allows the scaling up of problems onto as many processors as are available. This allows models to become more complex, and/or run at higher resolution with relative ease as more computer power becomes available. Sophisticated 3-dimensional ecosystem dynamics can only be computed on high performance computers and is ideally suited to parallel processors.

POLCOMS is formulated to run on parallel computers. Efficient communications are handled using the MPI (Message Passing Interface) standard. A sophisticated pre-processor allows total scalability to suit any machine, from a single processor workstation through to clusters (e.g. the POL 192-processor Linux cluster) to the largest machine available to the UK academic community, the 2560 processor IBM Power 5 machine at Daresbury Laboratory.

Scaling for multi-processors is done by 'domain-decomposition', where the model domain is split up into 'chunks' and each processor works on a chunk, communicating with surrounding processors for necessary data. Our decomposition algorithm maintains high load balancing for any number of processors.	Here is the decomposition for 256 processors.
The modular form of POLCOMS makes it easy to include additional processes or parameters.