多核乐观并行仿真的负载均衡研究

Load Balancing for Optimistic Parallel Simulation on Multi-core Platform

对于运行于多核计算机、基于多线程实现的乐观并行仿真,虽然操作系统可对线程进行调度以平衡各个核的负载,但它无法控制各逻辑进程本地虚拟时钟的平衡推进。提出了多核乐观并行仿真的四层负载分配模型及一种静态划分与动态负载均衡相结合的负载均衡方案。静态划分使用Metis图划分包对模型实例进行划分;动态负载均衡优先调度本地虚拟时钟较小的逻辑进程以实现各逻辑进程的平衡推进,无须进行模型迁移,易于实现。通过一序列实验检验了所提出的负载均衡方案的有效性。

For the optimistic parallel simulation implemented through multi-threading programming on the multi-core computer,though the operating system could schedule the threads so as to balance the load among cores,it can＇t balance the local virtual time advancement of logical processes.A four-layer load distributing model for the optimistic parallel simulation on the multi-core platform and a load balancing scheme that combined both static partitioning and dynamic load balancing were proposed.The model instances were partitioned using a graph partitioning package called Metis in the static partitioning,while the logical processes with a lower local virtual time were given higher priority to be scheduled in the dynamic load balancing scheme.The dynamic load scheme need not migrate model instances,and is easier to implement.The effect of the proposed load balancing scheme was verified through a series of experiments.