Design And Analysis Of Scalable Video Streaming Systems

Despite the advancement in multimedia streaming technology, many multimedia applications are still face major challenges, including provision of Quality-of-Service (QoS), system scalability, limited resources, and cost. In this dissertation, we develop and analyze a new set of metrics based on two particular video streaming systems, namely: (1) Video-on-Demand (VOD) with video advertisements system and (2) Automated Video Surveillance System (AVS). We address the main issues in the design of commercial VOD systems: scalability and support of video advertisements. We develop a scalable delivery framework for streaming media content with video advertisements. The delivery framework combines the benefits of stream merging and periodic broadcasting. In addition, we propose new scheduling policies that are well-suited for the proposed delivery framework. We also propose a new prediction scheme of the ad viewing times, called Assign Closest Ad Completion Time (ACA). Moreover, we propose an enhanced business model, in which the revenue generated from advertisements is used to subsidize the price. Additionally, we investigate the support of targeted advertisements, whereby clients receive ads that are well-suited for their interests and needs. Furthermore, we provide the clients with the ability to select from multiple price options, each with an associate expected number of viewed ads. We provide detailed analysis of the proposed VOD system, considering realistic workload and a wide range of design parameters. In the second system, Automated Video Surveillance (AVS), we consider the system design for optimizing the subjects recognition probabilities. We focus on the management and the control of various Pan, Tilt, Zoom (PTZ) video cameras. In particular, we develop a camera management solution that provides the best tradeoff between the subject recognition probability and time complexity. We consider both subject grouping and clustering mechanisms. In subject grouping, we propose the Grid Based Grouping (GBG) and the Elevator Based P lanning (EBP) algorithms. In the clustering approach, we propose the (GBG) with Clustering (GBGC) and the EBP with Clustering (EBPC) algorithms. We characterize the impact of various factors on recognition probability. These factors include resolution, pose and zoom-distance noise. We provide detailed analysis of the camera management solution, considering realistic workload and system design parameters