Monte Carlo criticality calculations using a parallel processor computer

The purpose of this work is to implement and evaluate the Monte Carlo criticality code KENO V.a on a parallel processing computer, specifically the NCUBE with 64 processors. Because the particle histories are independent of each other, it should be possible to track groups of particles simultaneously via the parallel computer and thus effect a savings in time relative to following the particles sequentially on a large mainframe computer. Several different versions of KENO V.a have been developed for the parallel computer starting with only the tracking subroutine and then progressively adding more subroutines for the parallel portion of the calculation. The results obtained indicate that the NCUBE is a fast computer only if nearly everything is put in parallel with little or no sequential calculations being performed. The banking of the histories into generations, an essential part of the Monte Carlo methodology employed in KENO V.a, must be done sequentially and, hence, slows the calculational process considerably. The problem caused by the banking process can be circumvented to some extent by using smaller banks of histories but this causes the statistical uncertainties to increase.

In summary, the 64 processor NCUBE is not ideally suited for conventional Monte Carlo criticality calculations. The host portion of the NCUBE computer, which performs the sequential part of the calculation, needs to be much faster. Alternatively, an interface between the NCUBE and a mainframe computer could be created, which is essentially a hybrid computer, to efficiently handle the parallel and sequential parts of the calculation, respectively.