The applicability of Monte Carlo codes to a parallel computer is studied. Two Monte Carlo codes-a radiation shielding analysis code MCACE and a criticality analysis code KENO-IV-are examined and modified in order to execute on an experimental highly parallel computer, AP-1000. Despite simple modifications without changing the original program flow, the MCACE code has achieved a speedup of 52 times on the AP-1000 equipped with 64 cell processors. In the case of the KENO-IV code, it is less suitable for parallel computers than the MCACE code, but a speedup of 13 times has been achieved by using 32 cell processors.
Two main causes of performance deterioration have been found; one is nonparallelizable part of the codes which affects more significantly with the increment of cell processors, and the other is idle times of cell processors which deteriorate the performance when the number of particles processed at each cell is small.
