Research Of Scheduling Algorithm In Web Cluster

Web applications and Http requests on Internet are increasing drastically recently. Accordingly many famous Web sites are facing the difficulty of server overload. Cluster technology is one of the effective methods of resolving this problem. On the other hand, with the popularity of the e-commerce, Internet's service mode is changing form traditional data communication and information browse to e-trade and service. Therefore the differential service performance must be provided and the QoS mechanism must be supported.Staring form the two aspects mentioned above, this paper focus on the issue that how to add QoS mechanism to the cluster circumstance so as to adapt modern web service requirement.Task scheduling and resource scheduling are two major aspects concerned with cluster system, while most of the arithmetic only consider one of them. In addition, recent research result shows that web access has a characteristic of self-similarity, so the influence of self-similar breaking-out task must be considered. To resolve those problems, an admission control-based QoS-aware dynamic request scheduling arithmetic is proposed in this paper. It introduces the probability Pa which combines self-similar network traffic character and act as a link to connect the task scheduling and resource scheduling. Pa is an function with two independent variables which are service class and load. With the increase of load, Pa varies non increasingly. For queens with different requirements, the Pa's varying speed is different so as to dynamically arrange bandwidth for every service class in the different load situations. At the end of this paper the imitate experiment is conducted to testify it's rationality.