Performance Modeling And Available Bandwidth Estimation In Wireless Ad Hoc Networks

Wireless ad hoc networks have drawn extensive attention and research interests,since they require no infrastructure and offer flexibility and convenience to users. Asone of the most fundamental and important performance metric, available bandwidth isa crucial factor in protocol design, quality of service (QoS) provision and networkmanagement, therefore, obtaining the information of available bandwidth is veryimportant in performance optimization in ad hoc networks. The traditional approachesof available bandwidth estimation that originated from wired networks cannot obtainreliable results when applied to wireless networks due to the unique characteristics ofwireless networks, whereas the approaches that are designed for wireless networkscannot ensure the validity of their estimation results since they ignore the interactionbetween the existing and incoming traffic. Consequently, this dissertation first buildsaccurate analytical models for single-hop and multihop ad hoc networks, and thendesigns the approach for available bandwidth estimation base on the models under theconstraints of overall QoS provision. The main works and contributions are as follows.We derive an accurate analytical model for IEEE802.11DCF under nonsaturatedcondition. The existing nonsaturated DCF models are obtained by scaling the saturatedones with the queue utilization, which ignore the characteristics of wireless networksunder nonsaturated condition and lead to inaccurate results. Consequently, we improvethe accuracy of the nonsaturated DCF model by considering the phenomenon of coupledqueues and the phenomenon of asynchronous backoff processes between wireless nodesunder nonsaturated condition. Based on the improved DCF model, we then present amodel for the statistical characteristics of service time, which forms the theoreticalfoundation of the queueing model. The advantage of the service time model lies in thatit distinguishes the influences brought by the successfully transmitted packets and thedropped packets and therefore eliminates the errors of existing models that abuse MACdelay as service time which will degrade the accuracy of the queueing model.We propose a principle and method for available bandwidth estimation based onoverall QoS provision. Existing researches on available bandwidth estimation lackexplicit definition on the availability of estimation results, which is the main cause oftheir invalid results. Consequently, we refine the definition of available bandwidthbased on the constraint of overall QoS provision, and propose a principle for availablebandwidth estimation under this constraint. Under our proposed principle, the feasibilityof bandwidth requirement is judged by checking whether the QoS requirements of theexisting traffic and the incoming traffic can be satisfied, and the available bandwidth isobtained by searching the maximum feasible bandwidth requirement. Our proposed isgeneric and can be applied to different kinds of wireless networks. We present an analytical model of single-hop ad hoc network for performanceanalysis, and propose an algorithm for its available bandwidth estimation. The featuresof our model lie in that (i) it can analyze the QoS metrics such as delay, throughput andpacket loss ratio;(ii) it takes into account the influences of nonsaturated condition,heterogeneous traffic source and queue length, and relates the network parameters to thebandwidth requirements of the flows; and (iii) it provides a what-if analysis, thusenables us to predict the performance of the network after a new flow with specificbandwidth requirement enters. Based on this model and the proposed estimationprinciple, algorithms for judging the feasibility of bandwidth requirement andestimating the available bandwidth are presented and validated through extensivesimulation experiments and shown to produce accurate results.We propose a MAC layer interference model and a link contention model inmultihop ad hoc networks. We first classify the interference according to its cause andeffect, and give the quantitive calculation for the interference brought by theneighboring nodes and the hidden nodes. This proposed interference model cancomprehensively describe the factors that influence the node’s behavior, and can reflectthe relationship between the extent of interference and the amount of load of thecompeting links. We then present an improved link contention graph to analyze theinterference between wireless links. Our proposed contention graph can qualitativelyanalyze the interference caused by both the hidden nodes and the neighboring nodes.Based on the qualitative analysis, the capacity variance problem of multihop path and itsbottleneck link problem are investigated and verified through extensive simulations.We present an analytical model of multihop path which takes into account theintra-flow contention problem and the dependency of loads between nodes, and researchon the capacity and available bandwidth estimation of multihop path. Based on theproposed MAC layer interference model, we model the activities of wireless nodesalong the multihop path, and derive the analytical model of multihop path forperformance analysis based on queueing networks theory and from which the QoSparameters of each flow along the path are obtained. The proposed model imposes noconstraints on the flow’s number and position, and is able to reflect the interactionbetween them. Based on this analytical model, we calculate the capacity of wirelesspath under the constraint of network stability, which overcomes the drawbacks ofexisting researches that ignore the state of the network when reaching the maximumthroughput point and hence improves the accuracy of capacity analysis. We also analyzethe available bandwidth of multihop path under the constraint of overall QoS provision,which further validates the generality of our proposed principle and method.