A new domain partition scheme for the everglades landscape model

This paper is based on a study of the domain partition problem for a given parallel/distributed application: the Everglades Landscape Model (ELM). The ELM is modelled using data parallelism and is implemented in a transputer system. A new partition strategy for load distribution in parallel/distributed systems is proposed. It ensures that each sub-area has approximately the same size and has at most four neighbouring sub-areas. This strategy has been used in solving the domain partition problem for the ELM. The result is compared with those from three previous partition methods: one dimensional partition, two-dimensional partition, and hierarchical partition. The comparison is based on several criteria proposed in our study. Results of the comparison show that the new partition method is the best among the existing ones.