Website Performance Evaluation and Estimation in an E-business Environment

Abstract

This thesis introduces a new Predictus-model for performance evaluation and estimation in a multi-layer website environment. The model is based on soft computing ideas, i.e. simulation and statistical analysis. The aim is to improve energy consumption of the website's hardware and investment efficiency and to avoid loss of availability. The aim of optimised exploitation is reduced energy and maintenance costs on the one hand and increased end-user satisfaction due to robust and stable web services on the other.

A method based on simulation of user requests is described. Instead of ordinary static parameter set, the dynamic extraction from previous log files is used. The distribution of existing requests is exploited to generate the actual based natural load. By loading the server system with valid and well-known requests, the behaviour of the server system is natural. The control back loop on the generation of work load assures the validity of the work load in the long-term.

A method for identifying the actual performance of the website is described. Using the well-known load in simulation of usage by a large number of virtual users and observing the utilisation rate of server resources ensure the best information for the internal state of the system. The disturbance of the service website usage can be avoided using the mathematical extrapolation method to reach the saturation point on the single server resource.