四类数值方法可扩展并行计算中动态负载平衡方法研究

This project is focused on the research on the dynamic load imbalancing problems arising from the scalable parallel computing applications including radial fluid dynamics Lagrange numerical methods, explosion fluid dynamics Euler numerical method, molecular dynamics, particle in cell method for laser plasma interation simulation, and dynamic Monte-Carlo method for neutron photon coupled transport etc., has successfully presented several efficient dynamic load balancing methods including multilevel averaging method, MPI parallel I/O, and domain decomposition indexed by physical model. In particular, by using the multilevel averaging dynamic load balancing method, this project has implemented the linear scalable parallelization of code Lared-I for laser driven internal explosion which is based on Lagrange numerical method, and has improved the performance of parallel molecular dynamics simulation by 25%. By using domain decomposion method indexed by physical model, this project has implemented the linear scalable parallelization of code MFIC for explosion fluid dynamics with Euler numerical method, and has obviously improved the parallel performance of code Lared-P for laser plasma interation simulation with particle in cell method. By using parallel MPI I/O, this project has implemented the scaleble parallelization of code P2DMC for simulation of neutron photon coupled transport dynamic Monte-Carlo method.

基于典型应用程序，揭示二维非定常流体力学Euler法和Lagrange法、二维电磁等离子体模饬Ｗ釉仆穹ā⒍肿佣ρЯＷ幽Ｄ夥椒ǖ人睦嗟湫褪捣椒衫┱共⑿屑扑阒校贾赂涸夭黄胶獾墓丶蛩兀⑻岢鲇行Ф涸仄胶夥椒ǎτ糜谑导饰侍猓夜咝阅懿⑿惺的Ｄ庥τ贸绦蚩⒌於崾祷。⑼贫夜⑿谢导视τ盟降姆⒄埂?..