集群服务器文件动态分配的改进算法

Improved Algorithm on File Dynamic Allocation of Cluster Server

在分析MMPacking算法的基础上,提出了一种改进的文件分配算法。在按照MMPacking算法分配文件时,根据节点的文件累积需求度去完成文件的分配或复制,考虑了节点的剩余能力。文件分配在所有服务器节点中周期性地进行,每进行了一轮文件分配后,都要从第1个节点开始新的一轮分配。在开始新一轮分配前,服务器节点要按照服务器的剩余能力重新进行降序排列。在每轮分配中,每分配一次文件到某个服务器节点后,都要检测当前节点服务器的剩余能力是否大于下一节点的剩余能力,如果满足条件,则将重新开始新的一轮文件分配。改进后的算法降低了由于客户需求或服务器配置变化所要支出的额外成本,有效地达到了负载均衡的目的。仿真结果表明,改进后的算法优于MMPacking算法。

An improved algorithm on file allocation among server clusters is proposed based on the MMPacking algorithm. File is allocated according to the degree of cumulative demand and the server＇s residual bandwidth. File is periodically allocated to all servers＇ nodes. After finishing each round of file allocation, the next file allocation begins with the first node. Before the file allocation of each new round, all nodes are arrayed again in descending order according to the new residual bandwidth. After a file having been allocated to a node, the new round of file allocation will be restarted if the residual bandwidth of CUlxent node is more excessive than that of the next node. This algorithm can reduce the extra system consumption that dues to the change of client demand or the change of server＇s configuration, and the load balance can be achieved more efficiently. Simulation results show that the improved algorithm is superior to the MMPacking algorithm.