$$R^2$$-Tree: An Efficient Indexing Scheme for Server-Centric Data Center Networks

Abstract

Index plays a very important role in cloud storage systems, which can support efficient querying tasks for data-intensive applications. However, most of existing indexing schemes for data centers focus on one specific topology and cannot be migrated directly to the other networks. In this paper, based on the observation that server-centric data center networks (DCNs) are recursively defined, we propose pattern vector, which can formulate the server-centric topologies more generally and design \(R^2\)-Tree, a scalable two-layer indexing scheme with a local R-Tree and a global R-Tree to support multi-dimensional query. To show the efficiency of \(R^2\)-Tree, we start from a case study for two-dimensional data. We use a layered global index to reduce the query scale by hierarchy and design a method called Mutex Particle Function (MPF) to determine the potential indexing range. MPF helps to balance the workload and reduce routing cost greatly. Then, we extend \(R^2\)-Tree indexing scheme to handle high-dimensional data query efficiently based on the topology feature. Finally, we demonstrate the superior performance of \(R^2\)-Tree in three typical server-centric DCNs on Amazon’s EC2 platform and validate its efficiency.