Video traffic modeling using Kolmogorov Smirnov analysis in Broadband Network

• Published in: Journal of Telecommunication, Electronic and Computer Engineering
• Main Author: Kassim M.; Samsuri R.; Yusof M.I.; Rahman R.A.; Isa N.M.; Mohamad R.; Ismail M.
• Format: Article
• Language: English
• Published: Universiti Teknikal Malaysia Melaka 2016
• Online Access: https://www.scopus.com/inward/record.uri?eid=2-s2.0-84999288615&partnerID=40&md5=992fc85ee8d8ccbef8b9e75688e6985e

Video Traffic utilization is one of the major issues for Quality of Service (QoS) for network traffic especially in broadband network. Most network administrators are looking at providing best QoS and reliable traffic performances especially on video traffic. Analysis on recent trend and modeling video traffic activity is a crucial task in providing better bandwidth usage. This research presents an analysis on video network traffic in a Broadband Network in Malaysia. Real data from a telecommunications service company based for Business and Home network are collected. Traffic characterization is analyzed and new traffic parameters and model are presented. Goodness of fit (GoF) and Kolmogorov Smirnov (KS) test is used to fit the real traffic in getting the best Traffic distribution model. Results present four top video used in the network traffic which are You Tube, MPEG, TV on Streamyx and Dailymotion using standard video protocol. Fitted traffics presents Pareto model is best fitted on video traffic. Generalized Pareto (GP) with Empirical Cumulative Distribution function (CDF) distribution is identified as the best distribution model. The fitted Generalized Pareto model was identified based on lower Kolmogorov-Smirnov (KS) value and higher probability value (p-value). Test statistics for four particular distribution results at 5% level significance. GP characterization presents three important parameters which are shape, scale and location. A new mathematical formulation is derived based on control parameters gathered for future rate limiting algorithms.