Measurement-based admission control with aggregate traffic envelopes

The goal of admission control is to support the quality-of-service demands of real-time applications via resource reservation. In this paper, we intro duce a new approach to measurement-based admission control for multiclass n etworks with link sharing. We employ adaptive and measurement-based maximal rate envelopes of the aggregate traffic flow to provide a general and accu rate traffic characterization that captures its temporal correlation as wel l as the available statistical multiplexing gain. In estimating application s' future performance, we introduce the notion of a schedulability confiden ce level which describes the uncertainty of the measurement-based "predicti on" and reflects temporal variations in the measured envelope. We then devi se techniques to control loss probability for a buffered multiplexer servic ing heterogeneous and bursty traffic flows, even in the regime of a moderat e number of traffic flows, which is important in link-sharing environments. Finally, we have developed an implementation of the scheme on a prototype router and performed a testbed measurement study, which together with exten sive trace-driven simulations illustrates the effectiveness of the approach in practical scenarios.