The QoS Research Of Web Cluster Based On Process Migration

Fast developing internet services induce more and more serious overload problem of servers directly. Web server cluster is a good solution with high performance, cost/effective ration for this problem. The key point of this solution is how to guarantee QoS. The core issue of QoS guarantee is how to shorten response time and utilize system resources effectively. Load balancing is very important approach of utilizing node's system resources, and core component of QoS control in application level. The loads among all the nodes will be adjusted dynamically by using load balancing in the environment of high density CPU consumption applications. It will provide the capacity of fast response duration accordingly. Therefor, we built a prototype server cluster system that based on preemptive process migration method by using extended linux kernel. Our purpose is to study method of QoS guarantee in cluster system that based on process migration.The thesis briefly introduce the background and current research status of Web server cluster , the key object and main content of Web QoS, thus get a conclusion on the necessary of the research on Web server cluster QoS.Aiming at the QoS control on application level and emphasis on solving load balancing issue with its policy and algorithm, the thesis has a overall discuss on cluster structure and its specific application environment, set up the high-efficient and reasonable topology of cluster. On the other hand, it also investigates and proposes solution on how to select appropriate target node and process, and get rid of the dated load information. In the thesis, we have illustration on the key technology to realize the software mode and the process status and ELF file format, etc, as well as the discussion on scheme of process interruption, migration and restoration. Finally, we did test and the analysis the system, the conclusion is QoS, which based on process migration can shorten the respond time for Web server cluster and guarantee the cluster's QoS.