基于对象存储系统中的热点数据平衡策略

Hotspot data balancing in OBS

针对基于对象存储系统(OBS)中的热点数据问题,提出了一种基于预测和数据副本迁移的均衡策略.对基于对象存储结点(OSD)的热度进行预测,元数据服务器(MDS)据此对存储结点进行均衡协调,存储结点则响应协调规则并采用贪心法选出尽量小而且访问客户多的对象进行副本迁移.此外,采用增量备份的方法解决副本迁移可能造成的数据不一致问题.实验结果表明策略的预测误差在8%以内.

A balancing strategy based on prediction and data copy′s migration was proposed in allusion to the hotspot data in OBS(object based storage).At first,it predicted the temperature of OSD(object based storage device),and then MDS(meta data service) coordinates them according to the states in prediction;OSDs executed the coordination,and migrate the copies of the object which is as small in size and more in accounts of related clients as possible,the object was chosen by the algorithm of Greedy.Moreover,the wa...