Simple and Efficient Parallelization Method for MOC Calculation

概要

論文の詳細を見る
A parallelization method that does not entail much data communication and allows easy implementation was developed for the method of characteristics (MOC). In the parallelization method, azimuthal angles were grouped to compute, on the same processor, the angular flux before and after the reflection on the outer boundary. The method can be easily applied to existing MOC codes without the highly technical knowledge on efficient data communication in parallel computing. It was implemented into the GALAXY code and numerical comparisons were performed. As a result, computation speedup by a factor of 9 to 10 and good parallel efficiencies of 70 to 80% were achieved by using twelve processors. The speedup benefits the practical reactor core design using MOC codes. Other than the parallel efficiency, the proposed method allows easy implementation because no drastic change in the conventional algorithm for existing sequential process codes is needed. This is carried out with the absence of data communication between processors during the inner iteration of MOC with the proposed method.
2010-01-01