基于多层油藏问题负载均衡的并行任务划分

Parallel Tasks Partitioning of Load-balancing for a Multiple Layers Numerical Simulation of Reservoir

该文基于分布式并行计算机系统，对一类多层二维二相流油藏数值模拟问题给出了３种任务划分策略──“卷帘”方式、区域分解方式和“卷帘”与区域分解结合的方式，对它们进行了比较，提出了减少求解时间、利于负载均衡和提高并行性能的任务划分方法，并实际应用于有多达７２万个网格节点的大规模油藏模拟问题实算结果表明，该策略划分产生的并行求解任务均衡，有利于加速比的提高．该方法也适用于区域或数据并行的任务划分问题.

in this paper, three tasks division strategies of load-balancing are given, which include a round-robinfashion, a domain partitioning fashion and a fashion of round-robin combining with domain partitioning, and acomparison is made. Then tasks division methods are proposed which can reduce solving times and benefit load-balancing and increases parallel efficiency for a simulation of reservoir of multiple layers two-dimension two-phase flow numerical problems on distributed memory parallel systems. This approach is applied to a large scalereservoir simulation which has over 720 000 grid points. The practical results show that the tasks generated by thestlategy are well-balanced and benefit to improve speedup. This approach is apt to the problems for domainparallel or data parallel.