一种基于跳跃hash的对象分布算法

摘要

如何有效地将海量数据分布到存储节点,是存储系统首要解决的问题.提出的MJHAR(matrix-based jump hash algorithm for replication data)对象分布算法简洁、高效,支持权值和数据冗余机制.该算法创造性地将节点映射到二维矩阵,对象的分布、定位只需从矩阵的行内、行间计算目标节点的行号和列号即可.理论研究表明,该算法满足公平性、自适应性、紧凑性、节点变化对象迁移量较小的特点.实验结果表明,该算法的计算时间比一致性hash算法快40％,比跳跃hash算法快23％,极大地缩短了计算时间,且比一致性hash算法对象分布更加均匀.%Efficient and scalable object management is a major factor of the overall performance in the object-based storage system.This paper proposes an algorithm MJHAR (matrix-based jump hash algorithm for replication data) that can be applied to the heterogeneous storage system to support variable levels of the object replication.The algorithm runs very fast and moves more minimizing data when storage nodes changes.Experimental results show that the running time of MJHAR is less than that of the consistent hash by 40％ and that of jump hash by 23％.The object distribution of MJHAR is more uniform than that of consistent hash algorithm.