Efficient Update Methods For Multi-dimension Metadata Indexing In Storage Systems

With the rapid growth of available information, storage capacity of large-scale storage systems has exceeded from the previous TB to PB or even EB scales. Searching and processing files is becoming increasingly difficult. Existing data storage systems based on hierarchical tree structure cannot effectively satisfy the requirements for the rapid growth of large-scale file systems and complexity of metadata queries due to poor scalability and functionality.Effectively organize metadata of storage systems, provide a variety of complex query services to help clients make better use of storage systems. This paper uses an R-tree with Bloom Filter to index multidimensional metadata, and support complex metadata queries. As existing researches on the R-tree mostly consider static data and are only concerned with efficient query processing, updating performance of R-tree is low. While metadata require frequent updates in file storage systems, direct modifications to the index produce large amounts of disk operations and imbalance index structure. The main contributions in this paper include a new update strategy based on R-tree structure index, which operates a group of update data as a new version. This strategy reduces the number of update operations, settles hotspot issues of repeatedly update and uses multi-version structure to optimize the update algorithm by the locality feature of update data.Compared with former work, this update strategy significantly improves the update performance of the index structure, reduces the time cost of update operations; the same time the efficiency of point query is also improved, while the efficiency of range queries has a slight drop.