A dynamic method for metadata partitioning based on intensive access of spatial data

In the object-based storage architecture of spatial data, accesses to metadata are 50% to 80% of the total data accesses, so metadata management and partitioning are very important. However, many typical and traditional methods for metadata partitioning, such as directory subtree partitioning, hashing partitioning, etc., should face the issues of hotspot and load balancing. In this paper, we analyzed the accesses to spatial data that follows Zipf-like distribution and has locality of reference, and proposed a dynamic method for metadata partitioning based on intensive access pattern of spatial data. This method considered the temporal locality and spatial locality of accesses to tile, put forward tile access rank algorithm based on the sum of access times per interval time, and got tile access probability by Zipf-like's law for dynamic hashing partitioning of metadata. The experiment results presented the improving of efficiency in tile access rank, and showed that the method for metadata partitioning is an effective solution for hotspot and load balancing issues.