A study on grid policy administrator with MDP-based resource management scheme

Abstract

Grid computing enables the virtualization of distributed computing and data resources such as processing, network bandwidth and storage capacity to create a single system image, granting users and applications seamless access to vast it capabilities. It brings together all heterogeneous resources and allocates them efficiently to applications. However, the complexity of resource management in Grid significantly increases as the grid computing grows. There are some approaches that try to solve the problem. One of the promising is policy based resource management system that is suitable to the complex management environments. In this thesis, we propose grid policy administrator with MDP-based resource management scheme. The grid policy administrator has a QoS-constrained grid resource management architecture based on Quorum and MDP-based policy decision scheme for considering management cost and QoS guarantee. The management architecture has two phase operations. The first complex decision procedure in these operations runs in proactive way to reduce brokering time. Then the second reactive procedure triggered by on-demand resource requests is served in time. Quorum systems are well known tools for ensuring the consistency of resource information as well as reliable QoS guarantee. Its useful properties such as intersection and mutual exclusion are used to make a large set of distributed heterogeneous resources several handy groups. Available resource quorums are generated by preplanned policy decision scheme with various SLAs, grid resource and network resource information in advance. When a request arrives those available resources are mapped into QoS quorums which satisfy QoS requirements. To evaluate the preplanned policy decision scheme with policy enforcement into network we run virtual heart simulation in our experiments. The results show that preplanned policy decision scheme guarantees QoS for grid applications.