针对本地分布式存储系统的拓扑结构与存储组成单元的特性,提出了一种基于频度的动态副本算法FBDR。FBDR分别采用基于密度的单次频度分析和基于区间长度的两次频度聚合的方法对文件访问流进行分析,作为确定热点文件的依据,具有较高的命中率。在副本创建位置选择上,综合考虑了存储单元的可用空间、负载、IO性能等因素,使热点文件获得更高的IO速率,同时兼顾了存储单元之间的负载与资源利用的平衡。
Based on the characteristics of topology and storage nodes of local distributed storage systems,a new Frequency Based Dynamic Replica(FBDR) strategy is proposed.FBDR uses two newly proposed algorithms called OFMI(One way Frequency Measure by Intensity) and TFMS(Two way Frequency Measure by Span) to identify hot files based on their access record history to improve the hit rate of hot files.FBDR takes available space,system load and IO performance of storage nodes into account to decide the location to replicate files,which not only improves the IO performance of frequently accessed files,but also balances the system load and resource consumptions on different storage nodes.