A Data Management Framework for Secure and Dependable Data Grid Approved by Supervisory

Data grid enables the sharing and coordinated use of data from various resources and provides various services to fit the needs of high-performance distributed and data-intensive computing. However, access efficiency, availability, and security are among the most important challenges to be addressed. In this research, we use data partitioning schemes (secret sharing and erasure coding combines with key encryption) to provide security protection for data storage, and replication techniques to provide access efficiency and high availability. We incorporate the proposed schemes and techniques into a data management framework which would provide high access efficiency, high system availability, and strong security assurance for data storage and access in data grids. In the dissertation, we focus on replica placement for correlated data in data grids, including data correlation due to transaction access and data partitioning. The topology of the data grid is modeled as a three level cluster-based general graph and both internet and mobile users are considered. Consider the replica placement of data that are correlated due to transactions/requests. We develop cost models for accessing correlated data objects in data grid with and without the effect of vii disconnection. Decomposed optimal and heuristic expansion-shrinking algorithms are developed for replica placement in mobile environment. Then, the solutions are extended to a tree topology and distributed optimal and heuristic solutions are provided. Consider the allocation problem for secure data objects that are partitioned and replicated. We first consider minimizing the access cost. The optimal replica allocation problem is decomposed into two sub-problems. The Optimal Inter-cluster Resident Set Problem (OIRSP) that determines which clusters need to maintain share replicas. The Optimal Intra-cluster Share Allocation Problem (OISAP) that determines the number of share replicas needed in a cluster and their placements. We develop heuristic algorithms for the two sub-problems. Then, we investigate the allocation problem for secure data objects to achieve multiple objectives, such as security assurance, system availability, and access cost. Models of security, availability, and access cost are developed to guide share replica allocation to achieve these objectives. Since these objectives may conflict with each other, the problem is modeled as a multiobjective problem. A genetic algorithm has been used to obtain Pareto-optimal solutions, which computes good solutions that achieve these objectives.