Improving QoS through Tra c Smoothing

The burstiness of variable bit rate traac makes it diicult to simultaneously make eecient use of network resources and provide end-to-end performance guarantees. In this paper, we present a smoothing scheme that allows network clients to optimize their quality-or price-of-service to respectively reduce their end-to-end delay bound or decrease their resource requirements from the network. Whether or not smoothing results in a net beneet to the network client depends on the the network load, the number of hops traversed, the burstiness of the stream, and the desired end-to-end delay bound. Hence, our scheme tells network clients if they should smooth, and if so, how much they should smooth to achieve the maximum beneet. We rst derive the analytical foundations for the scheme, and then perform empirical investigations using traces of MPEG compressed video.