Design And Optimization Of Distributed Multi-Dimensional Indexing For Data Centers

Having been widely used in data management,data analysis and data retrieval applications,data centers and distributed storage systems occupy important position in data services.Constructing efficient multi-dimensional index can effectively promote the performance of data centers and distributed storage systems,which is also a crucial strategy to guarantee data service quality.In this paper,we analyze the features of the R-Tree index structure and Fat-Tree Data Center Network(DCN).Base on the two-layer indexing framework,we construct FR-Index,a distributed multi-dimensional indexing schema for data centers.FR-Index finds the potential effect of DCN in data management,then it combines the advantage of R-Tree index structure to build a two-layer indexing system including several index instances in the data center.In the lower layer of the system,each server builds an R-Tree as local index to index the local data.In the upper layer,a distributed global index which is composed of a set of local index nodes depicts an overview of all data stored in the data center.Based on the layered structure of FR-Index,we design strategies to apply FR-Index to facilitate query processing.Considering the index update requirements,we provide different index updating strategies and propose index node cost model as the basis of those strategies.To promote the efficiency of index constructing and query processing,we provide several novel optimizing designs,such as constructing indexing spaces and dimension-ordered spreading algorithm.These optimizing designs fully take advantage of the architecture of the data center.We conduct performance evaluation for FR-Index on Amazon EC2 platform.The result shows that FR-Index is scalable and lightweight,which can process query requests efficiently.Comparison tests show that FR-Index behaves better on the data center than RT-CAN,an existing two-layer indexing schema.