VOD服务器集群中的改进SLF存储调度策略

Improved Smallest Load First Storage Scheduling Policy for VOD Server Clusters

在VOD服务器集群中,存储调度策略是影响整个系统存储容量和总并发数的关键技术之一.针对现有存储调度策略中最小负载优先(SLF)副本放置算法调整代价过高的问题,提出了一种改进SLF算法.算法以最小化负载不平衡度和最小化副本调整代价为目标,在放置过程中充分利用当前已经存储的副本,降低副本调整的代价.仿真实验表明,基于改进SLF算法的存储调度策略可以最小化负载不平衡度,降低了存储调度的调整代价,同时提高了系统的用户请求接受概率.

In clustered VOD （Video-On-Demand） server systems, storage scheduling policy greatly influences system＇s request scheduling, load balancing and scalability. However, each invocation of the existing smallest load first （SLF） replica placement algorithm is independent of the previous one, which could result in high cost of reorganization and make SLF impractical in many environments. To resolve this problem, an improved replicas placement algorithm is proposed. By making full use of previous stored replicas, the algorithm reduces the cost of adapting to changes in client demand or server configuration. Experiment shows that using the storage scheduling policy with proposed replica placement algorithm, our system achieves high user request acceptance rate as well as low load imbalance and low storage scheduling cost.