基于反馈的实时集群系统中的负载平衡方法

Load balancing method for feedback based on real-time clusters

为了提高实时集群系统中各节点的利用率并防止出现CPU使用率的热点,研究了实时集群系统中基于反馈控制的实时调度框架,并提出了一种新的面向任意图结构的负载平衡算法。该算法基于差异迁移系数和扩散负载平衡原理指导任务迁移以实现系统负载均衡,同时结合反馈控制以避免系统节点使用率振荡。实验结果表明,该算法不仅实现了节点间负载平衡、有效避免了局部热点,而且与反馈控制算法有机集成,保证整个系统稳定运行。

To raise the nodes＇ utilization ratio and avoid hotspot in real-time clusters, the feedback based real-time scheduling architecture is recognized, and a new graphic structure oriented load balancing method is proposed. Based on graphic structure, differentiated diffusive constant maker algorithm and diffusive load balancing algorithm is presented to guide tasks movement between neighbors. Besides that the feedback control algorithm is used to help avoid system oscillation. Simulation results show that the load balancing method works effectively under the architecture and have a robust performance.