The Research And Implementation On The Control Mechanism Of Executive Server In The Active Cluster System

With the rapid development of Internet and the wide use of web application, people need higher demand to Internet service than before. The web service is required not only to be high scaleable, high usable, easy management and high performance/price ratio, but also to provide with valuable functional features for some important applications, such as QoS control. All of these come up with higher challenges to web servers in the internet service platform.With the use of clusters, it is easy to achieve the goal of high usability and high performance. In addition, the performance/price ratio of clusters is very high. It has the unparalleled advantages. At present, most of clusters'scheduler polices adopt the centralized passive scheduler with the basic characteristic of centralized load evaluation, which mainly results in the inaccurate evaluation of server load. In the case of high system load, it is quite easy to have the"live lock"phenomenon for executive server, and thus reduces the system's efficiency. ASAS(Automatic Self-Allocating Server), invented by Shiyao Jin who is a professor of Computer Science Department in National University of Defense Technology, changes the former architecture by changing the passive mode into active mode. It distributes the load evaluation onto all executive server, which greatly decreasing the load of Dispatcher, improving the accurate of evaluation and preventing the occurrence of"live lock", thus making the huge improvement of clusters'overall performance.This thesis mainly focuses on the research of and development for capacity analysis software for executive server base on ASAS. Firstly, the thesis introduces the ASAS. Then algorithm for capacity plan and analysis is designed based on ASAS. The algorithm controls the capacity of all kinds of services on executive server, effectively makes use of the different kinds of resource usage information, such as network throughput, usage of CPU, usage of memory and so on. It guarantees the quality of service as well as making best use of the system resource, eliminating the phenomena of uncertainty of quality of service, unbalance of resource usage, and the occurrence of live lock. These offer a good foundation for QoS of Dispatcher. The following section in the thesis introduces how to trace the resources which fluctuates obviously and influences the quality of service clearly, such as network throughput, CPU use ratio and so on. Then it proposes the capacity plan and analysis algorithm which integrate the DMC auto control algorithm and fuzzy...