Indexing multi-dimensional data in modular data centers

Abstract

Providing efficient multi-dimensional indexing is critically important to improve the overall performance of the cloud storage system. To achieve efficient querying service, the indexing scheme should guarantee lower routing cost and less false positive. In this paper, we propose RB-Index, a distributed multi-dimensional indexing scheme in modular data centers with Bcube topology. RB-Index is a two-layer indexing scheme, which integrates Bcube-based routing protocol and R-tree-based indexing technology. In its lower layer, each server in the network indexes the local data with R-tree, while in the upper layer the global index is distributed across different servers in the network. Based on the characteristics of Bcube, we build several indexing spaces and propose the way to map servers into the indexing spaces. The dimension of these indexing spaces are dynamically selected according to both the data distribution and the query habit. Index construction and query algorithms are also introduced. We simulate a three-level Bcube to evaluate the efficiency of our indexing scheme and compare the performance of RB-Index with RT-CAN, a similar design in P2P network.