Monte Carlo simulation on transputer arrays (Q1098539)

A Monte Carlo simulation of a simple two-dimensional spin system as representative of a range of theoretical physics problems, is arranged for parallel computation by a multi-processor (transputer) array. Two essentially different types of decomposition; ``geometric'' where each processor handles a small sector of the physical system, and ``algorithmic'' where each processor handles a small sub-task within the full algorithm, are used and compared. The geometric decomposition is characterized by high efficiency in the utilization of processors and ease of programming, but the number of processors and the problem data- size are typically interrelated. The algorithmic method has greater flexibility of data size and makes minimal memory demands on most of the processors. In the particular example the algorithmic method was rather less efficient, but some of this relative loss is attributable to special features of the problem.