The Combination of Data and Functional Partitioning to Parallelize Severe Accident Computer Codes

Severe accident codes, i.e., codes that model core meltdown and accident progression in light water reactors, do not currently make use of parallel processing technology. Previous efforts to parallelize severe accident codes using DO-loop or data partitioning have resulted in speedup factors of <2.0 because of large serial code sections. Severe accident codes are more amenable to the functional partitioning approach, which splits a code into parallel tasks each representing a separate physical model. When combined, the two methods are able to partition 95% of the HECTR containment analysis code. Overall speedups of 2.6 and 3.2 on four and eight processors are obtained with the parallel HECTR code on an Alliant FX/80 parallel computer when modeling a moderately sized accident scenario. Speedups are expected to increase for larger severe accident codes, such as MELCOR, which contain more functional parallelism than the HECTR code.