电磁云计算的多目标任务调度算法研究

Research on multi-objective task scheduling algorithm for electromagnetic cloud computing

随着云计算在高性能计算领域的发展,大规模电磁有限元以及多样本计算逐渐采用云计算完成,因此云计算任务调度尤其是多目标任务调度成为一个需解决的重要问题。多目标任务调度算法优化了任务最大计算完成时间、机器总负荷和机器最大负荷3个目标。本文针对电磁有限元单样本计算不可分割的特性,提出多种群混合算法。采用多指标加权灰靶决策模型从Pareto解集当中选择最满意的云平台任务调度方案。在云平台上运行3个有限元计算案例,获得计算时间和资源消耗最优的CPU核心数和内存配置。通过测试基准和实际的案例,验证了算法的可行性和有效性,在云平台上实现了有限元高效计算和资源的充分利用。

With the development of cloud computing in the field of high-performance computing,large-scale electromagnetic finite element and multi-sample computing are gradually completed by cloud computing.Therefore,cloud computing task scheduling,especially multi-target task scheduling,has become an important issue to be solved.The multi-objective task scheduling algorithm optimizes the three objectives of maximum computing completion time,total machine load and maximum machine load.In this paper,Aiming at the indivisibility of electromagnetic finite element single sample computation,a multi-group hybrid algorithm is proposed.Making use of multi-attribute decision model based on weighted grey target to select the most satisfied cloud platform task scheduling solution.This paper obtains the optimal CPU core number and memory allocation by running three cases of finite element calculation on the cloud platform.The feasibility and effectiveness of the algorithm are verified by test benchmarks and actual cases.The finite element efficient calculation and the full utilization of resources are realized on the cloud platform.