Capacity analysis of internet servers based on log-data analysis

Due to the rapid increase in the internet traffic volume, ISPs are faced with the definite need of the expansion of server capacity. In order to provide prompt services for customers and still prevent excessive facility cost, it is critical to determine the optimum level of internet server capacity. The purpose of this paper is to provide a simple but effective strategy on the expansion of server capacity according to the increase in internet traffic. We model an internet server as an M/G/m/m queueing system and derive an efficient method to compute the loss probability which, in turn, is used as a basis to determine proper server capacity. The process of estimating the traffic parameter values at each server based on log data analysis is also given. All the procedures are numerically demonstrated through the process of analyzing actual log data collected from a game company.