The parallel processing of nuclear power plant severe accident simulation codes

This thesis describes the parallel processing of the nuclear source term code HECTR. The HECTR code simulates severe nuclear power plant accidents using detailed numerical models for relevant physical phenomena and plant safety systems.; The functional partitioning method is used to split HECTR into parallel tasks that each represent a different physical phenomena or safety system model. These models are solved in parallel over timesteps that are relatively short compared to the overall simulation time. The functional partitioning method is shown to be compatable with DO-loop or data partitioning. This new method allows a larger portion of source term codes to be solved in parallel, resulting in higher parallel speedups.; The parallel performance of HECTR is measured for three test problems of varying complexity and size. The parallel speedup of a typically sized problem is 2.32 on four processors and 2.63 on eight processors. The parallel efficiencies for these speedups are 58.1% and 32.9% respectively. Speedup is shown to be proportional to problem complexity and size. Adding functional partitioning to the data partitioned HECTR code is shown to result in a 34% speedup increase. This speedup is limited by the amount of available functional parallelism available in HECTR. Suggestions for further optimizations of parallel HECTR are made, and the potential speedup due to these optimizations is estimated to be as much as 10%.; The Speedup Improvement Equation (SIE) is derived to evaluate the effect on overall speedup of optimizations to specific parts of HECTR. The SIE is used to compute the benefit of selected optimizations in order to choose the best one for implementation.; The overall approach to parallelizing an existing serial code is discussed in detail. A recipe for this process is given, grouped into five main steps. Advice is also given for each step; this advice is discussed using specific examples from this research.