Voice over the Internet: Statistical properties and quality of service

Voice over Internet Protocol (VoIP) has been used increasingly to carry phone conversations. End-to-end VoIP is growing, but even more, service providers have been moving voice traffic to IP in their core networks; a voice call starts out on the traditional phone network, moves to IP in the core, and then moves back to the traditional network. Quality of Service (QoS) is a major issue in VoIP implementations, but is poorly understood. We are studying QoS using data collected from a real network, namely the Global Crossing Network (GBLX). Critical to the study of QoS is gaining a comprehensive understanding of the statistical properties of the important traffic components: call arrival process, call duration distribution, silence suppression, and packet arrival process. Every component involves different mathematical structures and requires a different style of statistical analysis. New visualization tools are developed to portray the dynamic details of every individual call, which are instrumental in validating the accuracy and consistency of the data and studying the call packet process. To estimate the power spectrum of the packet count process with changing mean and variance, a new statistical technique is developed to correct locally for the non-stationarity. Furthermore, to study QoS for scenarios that did not occur in our collected data, a queueing simulation is performed by varying the linkspeed and utilization rate