On the worst data loss behavior in statistical multiplexing

The introduction of QoS requirements into high-speed networks poses many technical challenges to network architects, for communication services usually request guarantees of performance bounds in a network. In the paper, the authors investigate the worst data loss behavior in a statistical multiplexer with limited buffer capacity. They evaluate the upper bounds on probabilities of buffer overflow in a multiplexer, as well as data loss of an input traffic stream. They found that simply controlling the utilization of a multiplexer does not improve the ability to guarantee performance. Indeed, these two performance parameters depend only on the available buffer capacity and the degree of correlation among the input traffic.