一种基于进程迁移的自适应双阈值动态负载平衡系统

Adaptive dual threshold dynamic load balancing system based on migrating

为了实现机群系统内负载的均衡分布 ,提高资源利用率和系统的吞吐率 ,论文设计并实现了基于并行虚拟机环境的可迁移动态负载平衡系统。该系统采用集中式任务调度 ,定义了负载向量的计算方法 ,同时设计了“自适应双阈值算法”实时监测系统下各结点的负载信息 ,并实现了同构机群系统下进程的动态迁移。该算法有效保证了负载信息的及时性和准确性 ,避免了进程迁移的颠簸现象。实验结果表明 ,由于采用了自适应算法 ,本系统额外开销较小 ,能有效改善动态任务的分配和调度 ,提高系统的并行运行性能。

Migratory dynamic load balancing system was designed to balance the workload in a network of workstations in order to enhance resource utility and increase system throughput. The system is a master slave scheduler and defines a workload vector in an adaptive dual threshold algorithm, which can detect the workload in each node of a network of workstations and can transfer processes between workstations on the homogeneous system. This algorithm results in a dynamically balanced workload and avoids excessive transfer. Test results show that the extra overload in the algorithm is small and it can effectively assign and reallocate tasks to improve the parallel performance of the load balancing system.