Time based internet traffic policing and shaping with Weibull traffic model / Mohd Azrul Abdullah

Quality of Services for bandwidth management is a crucial task in computer network. Policing and Shaping algorithms are identified as one of the task in bandwidth control, especially in tele-traffic engineering and computer network. This thesis presents the analysis of internet traffic flows in a campus network. Real internet traffic for inbound and outbound throughput flow is collected and statistical analysis is measured to characterize the real live internet traffic parameters. Empirical cumulative distribution function (CDF) model is presented in evaluating the real traffic distribution. Anderson-Darling (AD) and Goodness of Fit test is used to identify the best fitted distribution model to the real data. Four traffic distribution which are normal, lognormal, Weibull and exponential distribution are fitted and derived. Analysis results present Weibull Distribution model is the best fitted model. Two important Weibull parameters which are shape and scale are measured. Based on the identified statistical parameters, a new Time Based Policing and Shaping algorithm is developed and simulated. Policing process drops traffic while shaping process delay traffic to the next time transmissions. Mathematical model to formulate the algorithms is derived. The new algorithms present the existence inbound traffic burst is policed at 1200 Mbyte which is maximum allowable bandwidth while for outbound traffic burst, the policing traffic is policed at 680 Mbyte utilizes the maximum allowable bandwidth. The result of implementing policing and shaping traffic shows the burst can be controlled and thus, reduce the traffic congestion in the network. Furthermore, the drop data from policing traffic process can be saved and transmitted at the next transmission. Besides the method of congestion control, varying the shape parameter of Weibull distribution also help to reduce the burst and improve the performance the internet traffic as a whole.