Parallel methods development for a new version of the KENO-V.a critical safety code

KENO-V.a is a widely used Monte Carlo code for performing standardized computer analyses of nuclear systems for licensing evaluations. The code is used primarily to determine the reactivity (i.e., how close to critical) of a fissile system. It was originally developed for a sequential single processor computer. The program is large with approximately 1.5 megabytes of source code (primarily in FORTRAN). Due to the statistical nature of Monte Carlo and the desire for small uncertainties, KENO-V.a may require very long computation times for some problems. In this dissertation, the initial development of a parallel version of KENO-V.a for the Kendall Square Research supercomputer (KSRl) located at ORNL is presented. The KSRl is a shared memory parallel computer with 64 tightly-interconnected processors. Seven different parallel algorithms have been developed for KENO-V.a on the KSRl. In order to evaluate the methodology employed in the new parallel versions of KENO-V.a, several test problems have been utilized including the fuel drain tank/flush tank cell of the molten salt research reactor, a small reflected sphere on a plexiglas collar, and a critical Light Water Reactor problem. The Light Water Reactor problem is the same problem used by Sutton to evaluate the parallel performance of the Monte Carlo Code Racer which was developed at Knolls Atomic Power Laboratory. The cross-section libraries provided in the SCALE package are used in the models. The reference code for the speedup evaluation is the optimized sequential version of KENO-V.a for a single processor on the KSRl. The reduction in the computation time provided by the parallel code is quite significant. Speedup factors ranging up to 34.2 are obtained for up to 56 processors. Also, the efficiency is around 85% for 24 processors. This dissertation describes the investigations performed thus far with parallel KENO-V.a. The results demonstrate that the new algorithms developed in this work can provide accurate results with significantly reduced computation times relative to the conventional sequential version of KENO-V.a.