Performance analysis and optimization of a distributed Video on Demand service

Video on Demand (VoD) services are very appealing these days. In this work, we discuss four distinct alternatives for the architecture of a VoD server and compare their performances under different conditions. We later used the best configuration found in our analysis to evaluate, using simulations, the performance of a distributed VoD service in an ATM network covering an area the size of a large city neighborhood. We also introduced optimizations to the system, such as anticipated delivery and retrieval of video blocks from neighbor clients. Our results indicate that a server-driven operation mode (i.e., based on cycles) is not the most appropriate choice for a variety of workloads, even when the layout is striped (a result that challenges the conventional wisdom in the field). Also, our optimization strategies increased significantly the number of clients served in the system, which in our study case represented a savings of approximately 33% in the hardware required for the deployment of the service.