Accurate communication models for task scheduling in multicomputers

Models for computations and message-passing communications are very important for scheduling applications onto multicomputer systems and estimating the parallel execution time. This paper presents new models for communication, scheduling and estimating the execution time. In the proposed models, it is assumed that each processor in the multicomputer system has a router. The processor can only receive data from and send data to the router sequentially while the router can receive data from and send data to the routers of other processors in parallel. The execution time of a given application on a given machine can be then accurately estimated by using the proposed models. The input parameters to the proposed models are determined by measuring some parameters experimentally. Experimental results on the nCUBE2 machine are given to show the correctness of the proposed models