Scheduling Techniques for Broadcasting Popular Media
A media-on-demand (MoD) system is a distributed network system where servers respond to requests by clients to playback, view, listen, or read various types of media. Most current MoD systems use unicast, where each client receives its own transmission from a server. This solution cannot scale up well for popular media that may have hundreds of thousands of requests over a short period of time. With highly popular media, a server can become overloaded, causing denial of requests or large delays.
For popular media, broadcasting is the ultimate scalable solution. In the talk I present new broadcasting schemes for highly loaded MoD systems. These schemes exploit new technologies, which already exist or are expected to be available in the near future, such as the multicast capabilities of the Internet, high receive bandwidth at clients, and fast, cheap, storage at clients. Given these new technologies, the challenge is to find the best way to schedule the medias on the multicast channels to provide the best quality of service.
Our schemes are based on a new abstract model of the broadcasting system - an abstraction that reflects new technologies, mathematically models the various QoS parameters, and can handle different data-types. For a given bandwidth, these broadcasting schemes guarantee the minimum possible start-up delay time for an uninterrupted playback. Our results improve the best known asymptotic results and outperform known results for practical parameters of the system.
Joint work with Amotz Bar-Noy and Richard Ladner.