How to Improve the Scalability of Read/Write Operations with Dynamic Reconfiguration of a Tree-Structured Coterie

Abstract

In large-scaled environments such as computing grids, data replication is used to permit a better bandwidth usage of the network. Nevertheless, high latency time exposes the replica management protocols to potential performance degradations. Two new protocols are presented in this paper in order to avoid this degradation. Their goal is to ameliorate the access performance of replicated data in computing grids. The two protocols permit the dynamic reconfiguration of a tree-structured coterie (D. Agrawal and A. E. Abbadi, 1991) in function of the load of the machines possessing the data replicas. Each of the protocols permits to apply a tree transformation. The elementary permutation can be performed by having the load information of a small group of machines possessing the copies, whereas a global permutation must have the load information of all the machines that possess the copies. The implementation and the evaluation of our algorithms have been based on the existing atomic read/write service of (N. A. Lynch and A. A. Shvartsman, 1997). We demonstrate that the permutations permit to ameliorate the system's throughput. The results of our simulation show that tree reconfiguration based on the elementary permutation is more efficient for a small number of copies. The global permutation scales well and is more efficient when there is a large number of replicas.