Development orf a parallel computation technique based on multiple processes and its application to large thermal-hydraulic computer programs

Abstract

The complexity of the models and correlations in the best estimate thermal-hydraulic codes is ever increasing. Accordingly, the thermal-hydraulic calculations for reactor design require ever increasing computation power and resources. The computing power enhancement has been achieved through the developments on the component manufacturing technologies as well as on the computer architecyture. Some of these improved features have been utilized to speed up the simulations of the thermal-hydraulic computer programs. In this study, the Process-Level Parllel Computation (PLPC) technique has been developed based on the multiple processes in the computer with multiple CPUs. Then, the PLPC technique is applied to existing large thermalhydraulic computer codes, RELAP5/MOD3 and COBRA-TF to investigate the possibility of reducing the wall clock time and combining the above two codes. The development of COBRA/RELAP5 shows that the PLPC technique is very useful in combining several thermal-hydraulic computer programs that have similar numerical scheme. The COBRA/RELAP5 code, the merged version of COBRA-TF and RELAP5/MPD3, was produced to combine the excellent features of the two codes and thus to produce an enhanced best estimate code. During the merging process of the two codes, the inherent features of each code are not degraded so that the user can utilize fully the features available in each code. Through various test simulations, the merging scheme and its implementation are proven to be valid. These two large codes can be executed in parallel mode to reduce the wall clock time. The multi-RELAP5, which can have up to five RELAP5 processes work coherently on a single user job, shows the speed-up ratio of up to 3.41 in the ideal case with perfect load share on the desk side workstation APOLLO/DN10000. For the simulation of the typical Westinghouse nuclear power plant, the speed-up ratios are 2.51 with 3 CPU and 2.67 with 4 CPU machines respectively. With the problem of p...