Simulation and data analysis of peer-to-peer traffic for live video streaming

Abstract: Evaluating and testing changes or configurations to peer-to-peer systems or even understanding their behaviour can be complicated. One approach is to simulate a large peer-to-peer system and visualise the results. In this master's thesis a study is performed to understand how an actual implementation of a hybrid peer-to-peer live video streaming system behaves and performs under different scenarios. The behaviour and performance of a hybrid live video streaming system consisting of an unstructured mesh-pull-based P2P network and a classic content delivery network solution is studied by simulating the system with different scenarios such as flash crowds and flash disconnects. The simulation system includes a network model taking latency and bandwidth into consideration. As expected the mesh-based system performed well under user churn. Although the system consisted of approximately 80% free-riders the utilisation of the content distribution network was reduced by 95% on average. The data analysis was successful in improving the system's overall performance. Furthermore, the visualisations and data analysis were used to understand the system's behaviour.