Wireless Ad Hoc and Sensor Networks

Admission Controller Design for High-Speed Networks 149an accurate estimate of the highly nonlinear network traffic. Our proposedbandwidth estimation scheme takes advantage of a simplified networkmodel in discrete time, calculates the current bandwidth usage online,estimates the future bandwidth requirement, and assigns the availablecapacity fairly to new sources while meeting the QoS. Note that all availableschemes reserve bandwidth equal to the peak cell rate (PCR) of thesource for real-time voice and video traffic. Unfortunately, reserving bandwidthequal to PCR wastes bandwidth and other network resources,lowering the network utilization. The proposed online bandwidth schemeis shown to address this limitation.Only estimating the bandwidth online will not suffice to admit newtraffic. In fact, the issue of potential congestion also has to be consideredif a new source is being admitted into the network. Therefore, for anadmission control scheme to perform reasonably, a congestion indicatorhas to be used to decide the impact of adding a new traffic source ontoa network. In the literature, a number of congestion control schemes havebeen proposed (Jain 1996, Bae and Suda 1991). Unfortunately, many arereactive in nature and do not prevent congestion. Therefore, the novelpredictive congestion control scheme in (Peng et al. 2006) is utilized togenerate a congestion indicator, and it is subsequently used in the admissioncontroller. Finally, the past values of buffer occupancy are used topredict the availability of network resources for the new sources. Theadmission controller uses the resource and the bandwidth availabilitiesalong with the congestion indicator flag to decide whether to admit orreject the new source.In this chapter, the buffer dynamics of the ingress node, where the trafficis being admitted into the network, is modeled as a nonlinear discretetimesystem (Jagannathan 2005). A novel adaptive bandwidth scheme isdeveloped to estimate the network traffic onto the links by minimizingthe buffer occupancy error. Any function approximator can be used toestimate the traffic, but for simplicity a multilayer neural-network-basedadaptive estimator is proposed. Tuning laws are provided for the estimatorparameters, and closed-loop convergence with network stability is provedusing a Lyapunov-based analysis when an online bandwidth estimationscheme is deployed. Using the estimated and buffered traffic along withthe required QoS, the bandwidth required for all sources to meet the QoSis calculated. It is shown that the proposed online bandwidth estimationscheme guarantees the desired QoS by accurately estimating the bandwidthrequired by the existing sources, even in the presence of boundednetwork traffic uncertainties. A novel adaptive admission controller architectureis presented in this chapter from Jagannathan (2005) that unifies thecongestion controller, a bandwidth estimation scheme, and the proposedrule-based discrete-event controller. The performance of the adaptiveadmission controller is compared using service delay, cell/packet losses,