Parallel implementation of a transport network model

Abstract

This paper describes the parallel implementation of a transport network model. A ‘Single-Program, Multiple Data’ (SPMD) paradigm is employed using a simple data decomposition approach where each processor runs the same program but acts on a different subset of the data. The objective is to reduce the execution time of the model. The computationally intensive part of the model is within the assignment and simulation section and therefore this section is parallelised and executed using 1, 2, 4, 8 and 16 processors. The convergence, accuracy and performance of the parallel model are then assessed and compared to the linear implementation. The results indicate a performance increase of over 8 for the parallelised module and a speed-up of 5 for the total model when the model is run using 16 processors. The efficiency, average parallelism and efficiency-execution time profile are also discussed. In the context of time savings with 16 processors compared with 1, the time saving on the IBM SP2 are of the order of 80%, and, compared to a linear implementation on a dual processor Intel machine are of the order of 86%.