| Abstract | This thesis presents generic analytical models of homogeneous clustered Wireless Sensor Networks (WSNs) with a centrally located Cluster Head (CH) coordinating cluster communication with the sink directly or through other intermediate nodes. The focus is to integrate performance and availability studies of WSNs in the presence of sensor nodes and channel failures and repair/replacement. The main purpose is to enhance improvement of WSN Quality of Service (QoS). Other research works also considered in this thesis include modelling of packet arrival distribution at the CH and intermediate nodes, and modelling of energy consumption at the sensor nodes.
An investigation and critical analysis of wireless sensor network architectures, energy conservation techniques and QoS requirements are performed in order to improve performance and availability of the network. Existing techniques used for performance evaluation of single and multi-server systems with several operative states are investigated and analysed in details. To begin with, existing approaches for independent (pure) performance modelling are critically analysed with highlights on merits and drawbacks. Similarly, pure availability modelling approaches are also analysed. Considering that pure performance models tend to be too optimistic and pure availability models are too conservative, performability, which is the integration
of performance and availability studies is used for the evaluation of the WSN models developed in this study. Two-dimensional Markov state space representations of the systems are used for performability modelling. Following critical analysis of the existing solution techniques, spectral expansion method and system of simultaneous linear equations are developed and used to solving the proposed models. To validate the results obtained with the two techniques, a discrete event simulation tool is explored.
In this research, open queuing networks are used to model the behaviour of the CH when subjected to streams of traffic from cluster nodes in addition to dynamics of operating in the various states. The research begins with a model of a CH with an infinite queue capacity subject to failures and repair/replacement. The model is developed progressively to consider bounded queue capacity systems, channel failures and sleep scheduling mechanisms for performability evaluation of WSNs. Using the developed models, various performance measures of the considered system including mean queue length, throughput, response time and blocking probability are evaluated. Finally, energy models considering mean power consumption in each of the possible operative states is developed. The resulting models are in turn employed for the evaluation of energy saving for the proposed case study model. Numerical solutions and discussions are presented for all the queuing models developed. Simulation is also performed in order to validate the accuracy of the results obtained.
In order to address issues of performance and availability of WSNs, current research present independent performance and availability studies. The concerns resulting from such studies have therefore remained unresolved over the years hence persistence poor system performance. The novelty of this research is a proposed integrated performance and availability modelling approach for WSNs meant to address challenges of independent studies. In addition, a novel methodology for modelling and evaluation of power consumption is also offered.
Proposed model results provide remarkable improvement on system performance and availability in addition to providing tools for further optimisation studies. A significant power saving is also observed from the proposed model results. In order to improve QoS for WSN, it is possible to improve the proposed models by incorporating priority queuing in a mixed traffic environment. A model of multi-server system is also appropriate for addressing traffic routing. It is also possible to extend the proposed energy model to consider other sleep scheduling mechanisms other than On-demand proposed herein. Analysis and classification of possible arrival distribution of WSN packets for various application environments would be a great idea for enabling robust scientific research. |
|---|