Measurement-based admission control and monitoring for statistical rate-latency guarantees

Real-time Internet applications, such as voice and video, suffer from performance degradation whenever network congestion occurs. Statistical QoS guarantees, which provide a minimum rate but allow for infrequent violations of service, have the potential for greatly improving the quality of these applications by mitigating the effect of network congestion. We investigate a measurement-based admission control and monitoring algorithm that provides a minimum guaranteed rate and maximum scheduling latency, called a rate-latency bound, to traffic classes with the proviso that the guarantee may be violated for a small fraction of time. The algorithm estimates the demand for resources online by measuring the actual rate of each traffic class for particular choices of scheduling latency, and estimates the QoS violation rate using model fitting and extrapolation. The generality of these measurements allows one to provide a single admission control algorithm that can be used with many proposed guaranteed-rate servers.