Improved Distributed Spanning Tree and Case Study for Replica Location Service

Abstract

In the traditional distributed spanning tree (DST), randomly selecting representatives makes some nodes become critical nodes, which increase the load of critical nodes and also reduce the fault tolerance of systems. To resolve this problem, an improved DST structure and a searching algorithm were proposed. In this paper, we describe a representative selection rule first, which provides a good load balance and fault tolerance. Next, we define the searching radius to limit the flooding and proof that the time complexity of the searching algorithm is constant level. Finally, we give a case study for replica location service in data grid. Through the experiments with 16 nodes and simulations on overlay networks with larger scale, we present the performances of the structure. The results show that the time complexities of self-adaptive algorithms are logarithmic level, and the performance of the algorithm with searching radius boundary is better than the DST searching algorithm. The improved DST structure can be deployed in data grid to provide a good scalability, load balance, fault tolerance and flexible searching for various applications.