Research On Autonomous Scheduling Algorithm Based On Multi-agent Syntem

With the rapid development of modern network, it becomes much more heterogeneous and extensive. Therefore, traditional network architecture, algorithm and optimizing strategy can not meet the requirement of the rapid development, especially in those of distributed system with large throughput. The problem is that in large scale network, computing nodes and storage nodes appear to be dispersed and heterogeneous,these lead to the difficulty of common algorithm, so a much more efficient allocation strategy is needed. In addition, large scale networking computing system progress toward to be commercial, the needs of users tend to be heterogeneous, complex, so the system should be more time-efficient, reliable and secure; the network service provider should also reduce their cost and promote their efficiency. Therefore, the research on multi-performance optimization and related algorithm of large scale network system is very necessary.To solve the problems above, this thesis suggests an autonomous scheduling algorithm based on Multi-Agent system. The algorithm considers timeliness and reliability of the task, and then divides the task and allocates the task to the computing nodes. Universal generating function and simulated annealing algorithm are used to solve the problem. At last, a simulation is presented to verify the correctness of the model the thesis presented.Based on the former studies, the thesis integrates large scale computing system and Multi-Agent system together to build a large scale autonomous computing system and then, presents an algorithm of task division and computing nodes allocation. While the system gets the tasks from the users, it would analyze the tasks to make sure that whether they are best time or lowest reliability. After the Analysis, the system will divide the task into several small, parallel EB(executable blocks) with low coupling factor, and then allocate them to computing nodes for execution.The thesis also presents a mathematical model to evaluate performance of the system. The performance is time and reliability. A universal generating function would be use to analyze the performance.To solve the optimization problem the thesis presented, simulated annealing algorithm can be used. At last, the thesis presents a simulation to verify the correctnessof model. The two principles are all verified.