Measurement-based Per-flow Statistical Delay Guarantees

Admission control algorithms that provide statistical delay guarantees to real-time flows are of considerable interest since they can exploit statistical multiplexing of traffic to achieve higher link utilization. In this paper, we address the issue of providing per-flow statistical delay guarantees to real-time flows that traverse a link serviced by a ratebased scheduler. In such a scenario, each flow can specify a limit on the fraction of traffic that exceeds its delay bound requirements. We present a novel measurement-based admission control technique, called Measured Probability algorithm. Our algorithm allows multiple flows, that have heterogeneous delay bound and delay violation probability requirements, to co-exist on a single shared link. The algorithm exploits the fact that the actual delay experienced by most packets of a real-time flow is usually far smaller than the worst-case delay bound dictated by the rate assigned to the flow. Flows that can tolerate occasional delay bound violations can be assigned fewer resources than their worstcase requirement. The algorithm dynamically measures the ratio of actual delay experienced to the worst-case delay for each packet traversing the link and constructs a cumulative distribution function (CDF) of this ratio. The CDF is then used in estimating flow resource requirements whenever a new flow joins the system. Preliminary simulations indicate that, in addition to providing per-flow statistical delay guarantees, our algorithm can potentially improve the number of admitted flows by a factor of 10 and the link utilization by a factor of 20 when delay violation probability as small as 10−6 is allowed.