STOCHASTIC MODELING AND TIME-TO-EVENT ANALYSIS OF VOIP TRAFFIC

Voice over IP (VoIP) systems are gaining increased popularity due to the cost effectiveness, ease of management, and enhanced features and capabilities. Both enterprises and carriers are deploying VoIP systems to replace their TDM-based legacy voice networks. However, the lack of engineering models for VoIP systems has been realized by many researchers, especially for large-scale networks. The purpose of traffic engineering is to minimize call blocking probability and maximize resource utilization. The current traffic engineering models are inherited from the legacy PSTN world, and these models fall short from capturing the characteristics of new traffic patterns. The objective of this research is to develop a traffic engineering model for modern VoIP networks. We studied the traffic on a large-scale VoIP network and collected several billions of call information. Our analysis shows that the traditional traffic engineering approach based on the Poisson call arrival process and exponential holding time fails to capture the modern telecommunication systems accurately. We developed a new framework for modeling call arrivals as a non-homogeneous Poisson process, and we further enhanced the model by providing a Gaussian approximation for the cases of heavy traffic condition on large-scale networks. In the second phase of the research, we followed a new time-to-event survival analysis approach to model call holding time as a generalized gamma distribution and we introduced a Call Cease Rate function to model the call durations. The modeling and statistical work of the Call Arrival model and the Call Holding Time model is constructed, verified and validated using hundreds of millions of real call information collected from an operational VoIP carrier network. The traffic data is a mixture of residential, business, and wireless traffic. Therefore, our proposed models can be applied to any modern telecommunication system. We also conducted sensitivity analysis of model parameters and performed statistical tests on the robustness of the models’ assumptions. We implemented the models in a new simulation-based traffic engineering system called VoIP Traffic Engineering Simulator (VSIM). Advanced statistical and stochastic techniques were used in building VSIM system. The core of VSIM is a simulation system that consists of two different simulation engines: the NHPP parametric simulation engine and the non-parametric simulation engine. In addition, VSIM provides several subsystems for traffic data collection, processing, statistical modeling, model parameter estimation, graph generation, and traffic prediction. VSIM is capable of extracting traffic data from a live VoIP network, processing and storing the extracted information, and then feeding it into one of the simulation engines which in turn provides resource optimization and quality of service reports