基于闭合最小图划分模型的多作业分配优化方法

Multi-job Assignment Optimization Approach Based on Closed Minimum Graph-partitioning Model

随着并行计算系统规模的增大和复杂度的提高,已有的多作业分配方式可能导致较长的通信延迟和严重的通信竞争。针对这一问题,提出了一种基于闭合最小图划分模型的多作业分配优化方法。该方法以最小化通信延迟和消除通信竞争为出发点,通过建立闭合最小图划分模型,将多作业分配优化问题转化成闭合最小图划分问题,并设计闭合最小图划分算法来获得优化的多作业分配方案。

Due to the augment of the parallel computing system size and the increase of its complexity,the existing multi-job assignment approaches can cause severe communication latency and contention.In order to solve this problem,a new multi-job assignment approach based on a closed minimum graph-partitioning model was proposed.To minimize the communication latency and eliminate the communication contention,this approach translates the multi-job assignment optimization problem to the closed minimum graph-partitioning problem by building a closed minimum graph-partitioning model,and designs the closed minimum graph-partitioning algorithm to obtain an optimized multi-job assignment scheme.