Adaptive streaming techniques for large scale video delivery

The Internet has been a tremendous success in the past few years. New services offered by this data network are replacing or enhancing the traditional methods of communication. Multimedia applications such as distance learning, digital library, video conferencing are gaining more and more interest. However, despite increasing interest, these applications are not widely used. Such a situation can be attributed to a large degree to significant resource requirements of video streaming systems on which many multimedia applications rely.; A traditional video streaming system consists of a central server, a network and a client. Thus, the resources of interest include server I/O bandwidth, network bandwidth and client buffer space. We explore two directions to address bandwidth and storage space usage. The first direction relies on video caching by a proxy server and central server's ability to broadcast a video. A proxy server is located close to a client. It has the ability to cache part of the video and deliver it directly to the client. Periodic broadcast of a video allows us to reduce the bandwidth usage at the central server. Given that the proxy cache space is limited, we examine the effect of caching different portions of a video on the resource usage and construct server broadcast schemes for different types of caching. The analysis results in a group of algorithms that select a part of the video to be cached so that the bandwidth and client storage space are minimized, and the proxy storage space is utilized efficiently.; Next we define an architecture for video streaming on a large scale. We assume that proxy servers in different communities of clients may differ in their capacities, and that different communities may have different video access profiles. As a result, different proxy servers may choose different portions of the same video to cache. We explore methods to optimize the overall resource usage in such a heterogeneous environment. We include service replication in the architecture by introducing multiple geographically distributed central servers. We propose distribution of videos and client requests among these servers in a way that allows us to control the aggregate resource usage. We also investigate the problem of dynamic server selection that arises in a multiple server system.; The second direction in which we address resource requirements, explores adaptation of resource demands, namely network bandwidth needed to delivery a video, to the resource availability by controlling video transmission rate. We explore the possibility of performing rate adaptation in a network aware fashion while minimizing the effect it has on the quality of the video. The main challenge is in obtaining information about network. We propose to use a Multi-Level Explicit Congestion Notification-like mechanism to obtain the information about network backlog, construct a rate adjustment algorithm.