Abstract: Based on the analysis of the distribution of popularity, session length, load balance degree and other factors on system performance, a multi-objective replication and placement method in the streaming media cluster system is proposed. The problem of replica generation is summarized for the distribution of seats, and the optimal replication algorithm is proposed, which is heuristic to assign disk space for each replica of the streaming media replica to balance load, according to the characteristics of the file popularity of the Zipf-like distribution. On this basis, the multi-objective evolutionary algorithm is used for global optimal replica placement. Simulation results show that the method in the case of the storage space is limited to lower rejection rate and a higher load balance degree.

Key words: cluster, streaming media, replica generation, replica placement, load balance, reject rate

摘要： 分析流行度分布、会话长度、负载均衡度等因素对系统性能的影响，在流媒体集群系统中提出一种多目标复制存储方法。将副本生成问题归纳为席位分配问题，给出最优副本生成算法，根据文件流行度服从Zipf-like分布的特点，启发式地将磁盘空间分配给各流媒体副本，以平衡副本负载。在此基础上，设计基于进化的全局最优的副本放置算法。仿真结果证明，该方法可在存储空间受限的情况下实现较低的拒绝率和较高的负载均衡度。

关键词: 集群, 流媒体, 副本生成, 副本放置, 负载均衡, 拒绝率

CLC Number: 

• TP393