AN EFFICIENT TECHNIQUE FOR FRAGMENTATION AND ALLOCATION IN DISTRIBUTED DATABASE MANAGEMENT SYSTEM

Abstract

Distributed database provides the better solution to large-scale data management problems. A very important research issues is database system performance. Recently developed cloud database (CDBMS) is defined as distributed database. Similarly, in 2012, Bigdata is designed as a distributed database architecture running over 100s of clusters gi.e. machines. Distribution of data is a cumulative process of fragmentation, allocation and replication. The research objective of thesis is to propose an efficient technique for fragmentation and allocation in distributed database management system. The problem of fragmentation and allocation is considered as combined problem due to interdependency. However, replication is not considered in this research due to its high processing and communication cost as suggested by [4]. The allocation problem is NP-complete [3] and thus requires fast heuristics to provide efficient solution. Clustering of sites is done before fragmentation to ensure more efficiency, as clustering reduces the communication cost. The concept of clustering is proposed in [17] and fragmentation in [1]. Various strategies for allocation, such as cost based [4], preference based and nearest neighborhood allocation (NNA) [16], proposed are used to generate initial population for evolutionary algorithm developed. Finally the allocation of fragments is done using evolutionary algorithm developed in the thesis. Using the initial population of solutions, the algorithm proposed either generates a new solution for allocation or finds the best solution among the strategies considered. Much of attention has been paid to allocation as it is key factor in slashing query execution cost. Evolutionary algorithm is being developed due to its potential in solving NP- complete problems. The process of clustering, horizontal fragmentation, allocation strategies and evolutionary algorithm for allocation is taken as combined process and is applied to dataset of [1]. The system performance is enhanced using clustering at initial stages. Also, the performance of all the strategies and proposed algorithm is evaluated and the results show that the proposed algorithm provides near to the optimal solution for allocation of fragments to the clusters with the assumption of finding the average best solution.