Avaliação da conectividade: força, tendência e regularidade em redes oportunistas

Abstract

This thesis presents the development and validation of new routing protocols for Opportunistic Networks (OppNets). Routing in these networks is challenging due to the absence of end-to-end connectivity, requiring the store-carry-forward paradigm, where nodes store and physically transport messages until a contact opportunity arises. The core problem consists of maximizing delivery rates while seeking to reduce resource consumption (battery and buffer) in high-mobility scenarios. The first contribution is the CASTRO Protocol (Connectivity Assessment: Strength, Trend, and Regularity in Opportunistic Networks). Unlike approaches based solely on the frequency of encounters, CASTRO is based on the analysis of the complete interaction history, segmenting it into direct connection and disconnection intervals. From these data, the algorithm extracts three social indicators – Strength (S), Trend (T), and Regularity (R) – which together define node "Consistency" and "Scores", guiding controlled message diffusion. The second contribution is the QL-CASTRO Protocol, which integrates Reinforcement Learning (Q-Learning) into the CASTRO architecture. This model aims to mitigate the latency limitations of the base protocol by using dynamic delivery delay estimates instead of only social metrics. The evaluation methodology used the ONE simulator in two distinct scenarios: a dense pedestrian network based on the Helsinki map and a sparse vehicular network based on the road network of Manaus (AM), modeled using QGIS software. The results indicate that both protocols showed superior performance when compared to well-established and consolidated proposals (PRoPHET and Spray and Wait), as well as notable recent proposals (HESnW and QoN-ASW), reaching delivery rates close to 90% with moderate energy consumption.