Wireless Ad Hoc and Sensor Networks

148 Wireless Ad Hoc and Sensor Networksregulating the traffic flows according to changing network conditionsThis, however, requires an understanding of network dynamics. As aresult, it is not possible to determine the equivalent capacity of bufferthresholds for multimedia high-speed networks in bursty flow conditionsor under dynamic conditions (Dziong et al. 1997). On the other hand,conventional AC schemes, based on mathematical analysis, providerobust solutions for different kinds of traffic environments, but suffer fromestimation (modeling error) and approximation errors (real-time), so theyare not suitable for dynamic environments (Dziong et al. 1997). Networksare forced to make decisions based on incomplete information, which, ifnot done properly, degrades performance of the network. Therefore, aneuro-fuzzy approach for AC is proposed (Chen et al. 1999, Cheng et al.1999). Although this approach appears to make decisions in the presenceof incomplete information, it is not clear whether the approach can beimplemented in real-time, and further, no mathematical analysis is providedin Chen et al. (1999) and Cheng et al. (1999) to show its performance.Information on available bandwidth is required by the network todecide whether to accept a new source or not. Routes are typically selectedand sources accepted so as to minimize a certain measure of resourceswhile providing adequate QoS to the carried traffic. This requires anaccurate estimate of traffic conditions and the impact of adding a newsource. Subsequently, this information is provided for calculating theamount of bandwidth currently allocated to accommodate existingsources, and by identifying how much additional bandwidth needs to bereserved on links over which new traffic is to be routed.Because of the statistical multiplexing of traffic and shared buffers, bothaccounting and reservation are based on some aggregate statistical measuresmatching the overall traffic demand, rather than physically dedicatedbandwidth or buffer space per source. In order to obtain an accurateaccounting of bandwidth currently used, the current traffic onto theselinks has to be determined. The available bandwidth estimation schemesare based on simulation curves (Bae and Suda 1991). The major drawbackof using simulation curves is that in addition to requiring storage of theprecomputed curves at the bandwidth control points, the static nature ofthe information may not accurately reflect the dynamic and nonlinearbehavior of the network traffic conditions and connection characteristics.Although certain bandwidth estimation and allocation methods areheuristic (Jamin et al. 1997) in nature, others use neural and fuzzy logic(Chen et al. 1999, Cheng et al. 1999). Some of the schemes that utilize anestimate of the network traffic use the backpropagation algorithm or alinear least-squares method. Because the backpropagation algorithm isnot proved to converge in all circumstances and requires a learning phaseinitially, other online schemes were proposed. For instance, the linearleast-squares algorithm, though computationally simple, may not provide