Research On The Performance Evaluation Model Of Web Servers Based On Response Time Regression

Web servers play a dominant role in the service infrastructure of current Internet. It is their main task to receive and process requests of clients. In the current Internet the performance of service delivery crucially depends on the proper and efficient operation of Web servers. To ensure the ability of Internet services, it is necessary that Web server be evaluated correctly.Many popular Web sites and enterprises are facing huge concurrent connections or flows. This may slow response time, even bring down an entire server. We can simultaneously run multiple instances of different commodity operating systems with virtualization. Instead of grossly over-provisioning upfront due to uncertain demands, users can elastically provision their infrastructure resource from the provider's pool only when needed. Virtualization adds more power to the existing Internet technologies, but it also brings us a large challenge of evaluate the performance of Web server over virtual machines. In some senses, virtualization will reduce the performance of system. How to evaluate the performance of system over virtual machines becomes the hotspot in academic research and industry.In this paper, we aim at the research and analysis of models and methods of finding the peak rate which server can process. Then, we analyze the disadvantage of them and propose a Model which called MBR(Web Server Performance Evaluation Model Based on the Response Time). Further more, we implement the model in the virtual machine for benchmark testing. The main contents of this paper are listed as follows:(1) Traditionally, in order to get high accuracy they adopt same test number for every load. Now, we propose a new test method. By analyzing the relationship between the level and accuracy of loads, we find that different load level needs different test number to reach the same accuracy. When the load increases, we need more test number to reach specified accuracy. The test number dynamically changes, which can reduce the overhead cost.(2) Controlled Markov model is used to evaluate server performance according to the resource. This model is available only when the resource is bottleneck of server performance. In order to solve it, response time is used to guide the performance evaluation, which is a comprehensive refection of server performance. It can avoid considering the factors which will affect system performance.(3) Traditional test model which is based on response time, do not take the errors caused by network and so on into account. Moreover, they model only according to one or two times of experiment results. This leads to long time to let the results converge. Even more, the result evaluated by this method will bring some errors. In this paper, we consider the errors, and use regression method to deal with all of previous results, in order to get an accuracy results quickly.At present, virtualization is widely used in Web service. The research results of this paper have a theoretical innovation and value. In addition, MBRT is suitable for benchmark testing.