Research On The Controlling Policy Of Web QoS Based On The Request Classification

With the development of the computer network and multimedia technologies, the Internet is undergoing substantial changes from a communication and browsing infrastructure to a medium for conducting business and selling services. Most of the Web applications are the HTTP requests. Many web sites had to face the problem of the server overloading. The main symptom is that users have to wait a very long time for the HTTP response and even the HTTP requests will be rejected. The responding time is generally influenced by two elements: the quantity of the service of network's transfer layer and the performance of the web server. Recently years, the research has been carried out actively in this field. However, the Internet QoS mechanism cannot completely solve the end to end control problems. For example, in condition of the server overloading, if any measure was not taken about the QoS control in the applications layer,although the QoS of the network transferring is ensured very well, the request may also be rejected or the responding time be delayed so long that leads to "reject request" result. So we could conclude that the overload of the web server is becoming the botleneck of the QoS control. The Web server systems as core establishment must have the mechanisms and strategies to set up and sustain QoS, provide service differentiation and performance assurance to different users or HTTP request.Due to the above problems, in this article we do some key research in the field. We mainly focus on the strategy of classified services, the Qos control on the web servers, etc. The contents of the article are as following:(1) We study the mainstream technique and the architecture of Web QoS system, analyzing their kernel techniques. This work helps us master the latest developing trend of this field.(2) In this article, a Web QoS control strategy that is proposed is based on the session,the content of the request and the user's privilege. By the session,content of the request and the user's privilege, the customers can be classified into different classes. Different access control strategies can be imposed to the different classes customers. Higher-class customers can get more system resource, so that then can enjoy better service and faster HTTP response.(3) An Adaptive Priority Scheduling Policy is proposed to change the order of server process request. On the one hand, when the system was idle relatively, all requests which will be handled by the Web servers are from different queues. On the other hand, when the system was buzy relatively, the Web servers will set the scheduling order for the different queues, and then select the high-priority request to the Web server for processing.(4) An attribute reduction algorithm of rough set based on the minimal covering set is proposed. The attribute reduction in rough set is simplified the problem of the minimal covering set by constructing the relation matrix of the knowledge system.(5) A rough approximation-based clustering to cluster Web transactions form Web access logs is proposed. Based on the clustering analysis of the user's pattern, the different web pages will be combinated by their supportance. Using this approach, users can effectively mine web log records to discover web page access patterns. This is the basic theory of the claasification of user's request.