A Generic Scheme Based On Multiagent And Dynamic Redundancy In Load-Balancing System

Network of work stations (NOW) and workstation in Internet (WII) have both become efficient and feasible alternatives for traditional high performance distribute computing environment. But both NOW and WII are being faced up with the problem of effectively using the resources. So the issue of load balancing is vital to improving the performance of NOW and WII. Load-balancing can be classified as static, dynamic and hybrid. People are focusing their attention to dynamic load-balancing. Task scheduling in load-balancing algorithm can be classified as centralized and decentralized. In centralized load-balancing, a central scheduler to collect the global load information and make a scheduling scheme. While in decentralized load-balancing, scheduling scheme is made by schedulers in all the nodes together in the system. In a centralized load-balancing system, the central scheduler usually become the bottleneck. Many papers have studied the load-balancing schemes. But none of them have systematically taken brutal situations into account. A centralized load-balancing system is obviously sensitive to tough situations and the whole system will be wrecked once the central scheduler is down. The scheduler is bottleneck. Although in decentralized load-balancing systems, every node can act as scheduler, the global accuracy of scheduling will be affected if there exist nodes that are abnormal. So fault tolerance of load-balancing system is an issue that deserves our attention.The agent technology is playing a more and more important role in developing distributed system. More and more people have devoted to the challenging field and have achieved many results. The coming-up multi-agent system (MAS) greatly enhanced the efficiency and maintainability of distributed systems. And what is more important is that systems designed using MAS are closer to natural model, hence improve the adaptability of the system. So it is feasible to apply MAS in the designing of load-balancing system.In this paper, a generic security scheme for centralized load-balancing system, which utilizes the multi-agent technology to improve the communication between nodes in the system, is proposed. There are five kinds of agents in the system, which are monitoring agent, redundant-node selecting agent, master-slave communicating agent, executing agent and scheduling agent. Agents in each node of the system cooperate with each other so that all nodes can obtain the topology structure of the whole system. And the load information of the system can be acquired by the central scheduler timely and quickly, making the load-balancing decision more accurate. Scheduling decision is hence enhanced. The feasibility of using couple-computer as an approach to improve the fault tolerance is discussed in detail. A so-called dynamic redundancy is proposed, in the purpose of solving the bottleneck problem of the central scheduler. Dynamic redundancy is a technique based on couple-computer back-up and has the features of dynamic and high reliability. The dynamic of the scheme is based on the cooperation of the agents. The redundancy of the redundant scheduler is improved without affecting the usability of the whole system. Especially in tough environments, when the central scheduler tends to malfunction, the advantage of the scheme is notable. Because the scheme is generic, it can be used widely and has various methods to implement.