Implementation of a Monte Carlo code on a parallel computer system

Abstract

This paper reports the implementation of the EGS4 (Electron Gamma Shower version 4) Monte Carlo code system on a multiple-transputer system, the Edinburgh Concurrent Supercomputer (ECS). Basic aspects of running Monte Carlo calculations on parallel computer are described including distribution of random number seeds to the individual processes. Based on an independent task farm topology, an existing EGS4 usercode, DOSIMETER, which can be used to calculate the absorbed dose in cylindrical dosimeters placed in cubic phantoms, has been re-written for running on the ECS. The parallelised code consists of three sequential programmes: 1. (1) a ‘control’ process to generate tasks; 2. (2) a number of ‘simulation’ processes to consume tasks and to generate results; and 3. (3) an ‘analysis’ process to collect and to analyse results. A linear increase in computing speed has been achieved with the number of ‘simulation’ transputers used on the ECS. The basic techniques used in this work are also suitable for the implementation of the Monte Carlo code on other parallel computer systems which the implementation of the Monte Carlo code on other parallel computer systems which have similar operational structures as the ECS.