Aprimorando o desempenho de algoritmos de roteamento em VANETs utilizando classificação

• Autor(a) principal: Costa, Lourdes Patrícia Portugal Poma
• Data de Publicação: 2013
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da UFSCAR
• Texto Completo: https://repositorio.ufscar.br/handle/ufscar/544
• topic: Redes de computação; Redes tolerantes a atrasos e desconexões; Sistemas de comunicação móvel; Ad hoc networks (Redes de computação); Delay and Disruption Tolerant Networks; DTNs; Mobile Networks; Mobile Ad-Hoc Networks; MANETs; Vehicular Networks; VANETs; CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO

Vehicular ad-hoc networks (VANETs) are networks capable of establishing communications between vehicles and road-side units. VANETs could be employed in data transmission applications. However, due to vehicle mobility, VANETs present intermittent connectivity, making message transmission a challenging task. Due to the lack of an end-to-end connectivity, messages are forwarded from vehicle to vehicle and stored when it is not possible to retransmit. Additionally, in order to improve delivery probability, messages are replicated and disseminated over the network. However, message replication may cause high network overhead and resource usage. As result, considerable research e_ort has been devoted to develop algorithms for speci_c scenarios: low, moderate and high connectivity. Nevertheless, algorithms projected for scenarios with a speci_c connectivity lack the ability to adapt to situations with zones presenting diferent node density. This lack of adaptation may negatively a_ect the performance in application such as data transmission in cities. This masters project proposes develops a method to automatically adapt message replication routing algorithms to diferent node density scenarios. The proposed method is composed of three phases. The first phase collects data from message retransmission events using a standard routing algorithms. The second phase consists in training a decision tree classifier based on the collected data. Finally, in the third phase the trained classifier is used to determine whether a message should be retransmitted or not based on the local node density. Therefore, the proposed method allows routing algorithms to query the trained classifier to decide if a message should be retransmitted. The proposed method was evaluated with real movement traces in order to improve Spray and Wait and Epidemic routing algorithms. Results indicate that the proposed method may contribute to performance enhancement.