An efficient and scalable multi-dimensional indexing scheme for modular data centers

Abstract

An efficient distributed indexing scheme plays an important role in improving the performance of cloud storage systems. To achieve concurrent query service and high manageability, the indexing scheme should meet the requirements of high scalability and low latency. In this paper, we propose RB-Index, an efficient and scalable multi-dimensional indexing scheme for modular data centers with the BCube topology. RB-Index is a two-layer indexing scheme integrating the BCube based routing protocol and the R-tree based indexing structure. In RB-Index, we build several distinct indexing spaces with dimensions selected according to query history. Each server takes responsibility for a portion of the indexing space according to a mapping scheme. A data pretreatment method and a publishing scheme are presented to uniformly distribute the global index across all the servers in the network. Index maintenance strategies are designed to keep the system cost at a low level. Efficient and complete query strategies are also introduced to support highly concurrent queries. We conduct experiments on Amazon EC2 platform to evaluate the performance of RB-Index and compare its performance with RT-CAN and FT-Index. Experiment results manifest the efficiency and scalability of our indexing scheme.