Predição de encontros de nodos móveis para aperfeiçoamento da detecção de defeitos em manets

Abstract

A MANET (Mobile Ad hoc NETwork) is a high-strength and easy maintenance network of mobile nodes with ad hoc topology; besides, it is accessible anytime and anywhere, without the need for base stations. A Failure Detector (FD) is an “oracle” that uses messages exchanged on the network to identify faulty nodes. Since an FD can misidentify as faulty nodes, its suspicions are only used to prevent distributed systems from waiting indefinitely for a faulty node to respond. The mobility and auto-configuration characteristics of a MANET make this type of network sensitive to the problem of differentiating a node disconnection due to its motion from a node-failure, which requires the presence of an FD. Regarding node mobility detection, the models studied assume “passive mobility”, where nodes do move but have no notion of it, and consequently cannot notify their mobility. This lack of knowledge leads to scenarios where nodes can be mistakenly identified as faulty, as MANET nodes are unaware of the possible mobility trajectories of the monitored nodes. This work proposes the use of Machine Learning techniques to explore nodes, whose motion is not random, to implement a Meeting Map from the prediction of their average velocities over time, which is the result of the intersection of routes of different objects in the same time slot. By using this map, we can obtain optimal values for the FD timers. We have thus developed an FD, named 𝑀𝐴𝑓𝐷, which outperforms competitors and, as an unexpected result, produced an energy-saving efficiency method for opportunistic or Delay Tolerant Networks (DTNs).