Scalable reliable on-demand media streaming protocols
This thesis considers the problem of delivering streaming media, on-demand, to potentially large numbers of concurrent clients. The problem has motivated the development in prior work of scalable protocols based on multicast or broadcast. However, previous protocols do not allow clients to efficiently: 1) recover from packet loss; 2) share bandwidth fairly with competing flows; or 3) maximize the playback quality at the client for any given client reception rate characteristics. In this work, new protocols, namely Reliable Periodic Broadcast (RPB) and Reliable Bandwidth Skimming (RBS), are developed that efficiently recover from packet loss and achieve close to the best possible server bandwidth scalability for a given set of client characteristics. To share bandwidth fairly with competing traffic such as TCP, these protocols can employ the Vegas Multicast Rate Control (VMRC) protocol proposed in this work. The VMRC protocol exhibits TCP Vegas-like behavior. In comparison to prior rate control protocols, VMRC provides less oscillatory reception rates to clients, and operates without inducing packet loss when the bottleneck link is lightly loaded. The VMRC protocol incorporates a new technique for dynamically adjusting the TCP Vegas threshold parameters based on measured characteristics of the network. This technique implements fair sharing of network resources with other types of competing flows, including widely deployed versions of TCP such as TCP Reno. This fair sharing is not possible with the previously defined static Vegas threshold parameters. The RPB protocol is extended to efficiently support quality adaptation. The Optimized Heterogeneous Periodic Broadcast (HPB) is designed to support a range of client reception rates and efficiently support static quality adaptation by allowing clients to work-ahead before beginning playback to receive a media file of the desired quality. A dynamic quality adaptation technique is developed and evaluated which allows clients to achieve more uniform playback quality given time-varying client reception rates.
Doctor of Philosophy (Ph.D.)