高效的元数据索引是提高海量存储系统性能的重要手段.针对现有元数据管理方法存在的时间与空间开销大和性能不稳定等问题,我们设计了基于属性分频的元数据索引算法.依据元数据中属性被访问的频率等因素,分解元数据分别存储到高频元数据属性集和低频元数据属性集中,使用KD-tree建立高频元数据属性集的索引,满足多条件混合查询高频元数据属性的要求;使用人工免疫算法建立低频元数据属性集的索引,在保持较高查询性能的同时,避免大量额外的存储空间.实现了算法的原型系统,使用两个真实数据集进行了测试与分析,结果表明基于属性分频元数据索引算法具有时间与空间开销小、适应能力强的特性.
Efficient metadata indexing is important to improve the performance of mass storage system.Current metadata management algorithms need large and volatile time and space overhead,we present the attribute divider based metadata indexing algorithm.All metadata is divided into high frequency metadata attribute set and low frequency metadata attribute set.Using the KD-tree to index the high frequency metadata attribute set and using the artificial immune algorithm to establish the index of low frequency metadata attribute set.Then analyzing the attribute divider based metadata indexing algorithm from performance and the adaptability.Finally we realized the prototype of the attribute divider based metadata indexing algorithm,and used two real test data sets to test and analyze.The results show that the attribute divider based metadata indexing algorithm has low time and space overhead of metadata query,and high adaptability.