Performance modeling of web servers

A general model of a web server system comprising of the interactions between World Wide Web users and the web sites (servers) is analyzed and evaluated. Incoming requests, once admitted for processing, compete for the available resources (HTTP threads). An efficient approximate solution is provided; its accuracy is evaluated by comparing the model estimates with those obtained from simulations. The effect of several controllable parameters on the performance of the system is examined in a series of numerical and simulation experiments. In trying to understand the interactions between web users and web servers, we attempt to answer three key questions. How can we model user and server behavior on the World Wide Web ? How do users and web servers interact? Can we improve upon the ways in which web servers process incoming requests from users? In our study we formulate a queueing model for the web server and from the queueing model we obtain expressions for web server performance metrics such as average response time, throughput and blocking probability. This model will be used evaluate the suitability of web servers to prospective users of web server systems. The foreseen end users of the model are corporate decision makers who faced by a variety of several web server systems, are interested in evaluating the suitability of the servers in market. We envision a situation in which a given manager has a set of his/her own requirements or analysis of the business requirements and needs to purchase a web server that can meet the demands/requirements of the situation at hand. Hence with the users requirements and server specifications, the model could predict the best web server for the user requirements. We model the web server as an M/M/1/K queue with FCFS queueing discipline. The arrival process of HTTP requests is assumed to be Poissonian and the service discipline First come First served (FCFS). The distribution of service time is assumed to be exponential. The total number of requests that can be processed at one time is limited to K. We obtain closed form expressions for web server performance metrics such as average response time, throughput and blocking probability