Improving the routing layer of ad hoc networks through prediction techniques

Laburpena

Every day becomes more evident the key role that mobile computing and wireless technologies play in our daily activities. Being always connected, anytime, and anywhere is today more a necessity than a luxury. The ubiquitous computing scenarios created based on these technology advances allow people to provide and consume shared information. In these scenarios, the supporting communication networks are typically wireless and ad hoc. The dynamic and changing characteristics of the ad hoc networks, makes the work done by the routing layer to have a high impact on the performance of these networks. It is very important for the routing layer to quickly react to changes that happen, and even be advanced to what will happen in the near future, by applying prediction techniques. This allows the routing layer to have a proactive approach (try to avoid a problem before it happens), which is more efficient than a reactive approach (which try to correct the problem when it is already present).In this context, the general research question that this thesis investigates is What is the improvement achieved when we apply prediction to the routing layer of adhoc networks? In order for the routing layer to do its job (to decide what is the best path for an information to reach its destination) it is necessary to know what paths or links are available among the network nodes (topological information) and how good these paths or links are (quality). For this reason, the general research question of the thesis is addressed in two key aspects: (1) the prediction of the topological information, and (2) the prediction of the quality of ad hoc networks. This thesis investigates whether prediction techniques can improve the routing layer of ad hoc networks. As a first step in this direction, in this thesis we explored the potentiality of a History-Based Predictor (HBP) strategy to predict the Topology Control Information (TCI) generated by routing protocols. We demonstrated that there is a high opportunity for predicting theTCI, and this prediction can be just focused on a small subset of messages. Based on our findings we implemented the OLSR-HBP predictor and evaluated it with regard to the Optimized Link State Routing (OLSR) protocol. OLSR History-Based Predictor (OLSR-HBP) achieved important decreases of TCI (signaling overhead), without disturbing the network operation, and requiring a small and affordable amount of resources. Finally, regarding the impact of Prediction on the routing data for both Link and Path (or End-to-End) Quality information, we demonstrated that Time-series analysis is a promising approach to accurately predict both Link and End-to-End Quality in Community Networks.