Routing and traffic-engineering in multi-hop wireless networks: an optimization based approach

Multi-hop wireless networks (MHWNs) attract significant interest due to the minimal infrastructure demands and their potential in supporting mobile and pervasive computing. The high demand placed by a growing user base on the limited available bandwidth places a premium on effective communication and networking for MHWNs. Recent studies in protocol development have improved the network performance considerably. However, effective networking under sparse bandwidth remains a difficult problem. Traffic-engineering techniques have been used to solve this problem in conventional networks. One of the key challenges in traffic-engineering in MHWNs is the routing problem - a problem of delivering packets across multiple wireless hops, considering the complex interference patterns and interactions in MHWNs. The dissertation targets developing a formally grounded approach for solving routing problems in MHWNs, while taking into account the effects of interference. The work builds on recent efforts in the networking community to express a network behavior as an optimization problem, and decomposing the formulation to provide distributed protocols. A successful model can then be applied to: (1) analyze the performance and capacity of existing protocols; (2) develop protocols for traffic engineering and admission for static networks; and (3) develop formally grounded and near-optimal distributed routing protocols. In MHWNs, the routing problem is substantially more complicated than the wired problem because of interference. Interference is exhibited at many levels, leading to effects such as uncontrolled contention and unfairness. The approach proposed in this dissertation breaks the problem into multiple layers. Firstly, a Multi-commodity flow based routing model that produces interference-separated routes is developed. The study analyzes the interaction of multiple routes and proposes effective objective functions to improve the throughput and delay metrics of the connections. Simulation results show significant improvement in performance over a traditional routing protocol. However, the assumption of an ideal scheduling model in the routing model discounts for the scheduling effects, that limits the applicability of the model. Existing accurate CSMA scheduling models cannot be directly applied as part of a routing formulation due to their complexity and underlying assumptions. A low-complexity scheduling model is proposed to capture the key scheduling interactions in CSMA based schedulers, like IEEE 802.11, and this model is integrated with the routing model. Simulations demonstrate large performance improvements (around 40%) when compared to the scheduling-unaware routing model. The above routing models are NP-hard, thus limiting their applicability to larger networks. The third contribution of the dissertation is to approximate the routing model to a polynomial time algorithm. A decomposition based approach is followed to formulate a low-complexity model by applying domain-specific heuristics. We show that such a model is orders of magnitude faster than the above routing models, while being able to maintain efficient solutions. An approach to account for the effect of scheduling, while preserving a low run-time, is presented. Interaction graphs are proposed to capture the scheduling characteristics of CSMA based schedulers. Using the interaction graph framework, we capture a fundamental fairness property of CSMA scheduling called Contention fairness. The dissertation proposes a detailed throughput estimation model to capture contention fairness by applying Renewal Theory and Continuous-time Markov chain. Based on the insights gained, we propose a distributed mechanism to mitigate unfairness. Simulation results validate the accuracy of the model and demonstrate the significant improvement in fairness obtained bythe proposed distributed scheme. The next contribution of the dissertation is to characterize the interactions between two competing links in CSMA. Throughput estimation models are proposed for various categories of interactions that have been identified in MHWNs and are validated by simulations. We show that such a model can accurately capture the certain frequently occurring categories. We discuss the future work towards realizing the formulation as an online traffic-engineering tool in deployed networks. Finally, we sketch our plans towards formally deriving routing layer functionality in distributed protocols using the above framework.