Statistical multiplexing and buffer sharing in multimedia high-speed networks: a frequency-domain perspective

We study the effectiveness of statistical multiplexing and buffer sharing under the multimedia high-speed networking environment. We focus on the impact of frequency-domain source characteristics on dynamic resource sharing. By applying a novel statistical matching technique, we can, for the first time, investigate the multiplexing performance of a wide-range of realistic traffic sources using sophisticated traffic models. It has been shown that the effectiveness of statistical multiplexing and buffer sharing highly depends on the frequency-domain characteristics of the traffic as well as the corresponding QoS requirements. For practical "low-frequency" sources; e.g., VBR-video streams and LAN-to-LAN traffic, we show that significant savings in bandwidth and buffer-space can be achieved via resource sharing under practical loss and delay constraints. These findings re-illustrate the important role of traffic characteristics in the design/selection of network control strategies. The trade-offs among different design alternatives (e.g., multiplexing versus buffering) and the implications on some control schemes, e.g., traffic shaping/input-rate control, are also discussed.