Quality-of-service dependent traffic regulation for multimedia networks

In multimedia networks, certain applications (e.g., voice and video sessions) can tolerate some delay violations and packet losses. For these applications, we proposed a novel, more flexible traffic characterization approach-L^p-sense constraint functions-as a generalization of deterministic constraint function. Intuitively, feeding more packets than provided resources will degrade the quality-of-service (QoS) of multimedia applications. However, if we can control a multimedia application´s output to conform to constraint functions in L^p sense, then the portion of time prone to performance-bound violations has a normalized tail bound decaying with x ^-p. With the notion of L^p-sense traffic characterization, we investigate algorithms to shape an arbitrary arrival trace to conform to a given function f in L^p sense, i.e., L^p-sense regulation algorithms. It can be shown that under certain conditions LP-sense regulation departure times converge to deterministic regulation departure times when p→∞. Hence L ^p-sense regulators can be regarded as generalizations of the continuous-time optimal deterministic regulators. By properly choosing p, more packets of a multimedia source can pass through L^p-sense regulators while its normalized bound-violation ratio decays fast enough to meet its QoS requirements. The basic “leaky bucket” type L^p-sense regulators are also studied as an illustration