A distributed object-oriented framework for FDTD

Some aspects of the architecture of a distributed object-oriented application framework for the Finite Difference Time Domain (FDTD) algorithm are presented. The target architecture for the frameswork is a Beowulf-class cluster of PCs and SMPs with an inter-node API based on message passing semantics such as PVM and MPI and intra-node semantics on SMPs based on kernel threads and shared memory. Implicit in the choice of computer architecture is that computation must proceed at two levels of granularity simultaneously, coarse grained (task, process) across the distributed memory multicomputer and fine-grained (threads) across processors in SMPs. Construction of the run-time computational structure to support this approach is presented.