Um modelo baseado em reputação e protocolo de rumores para evitar ataques sybil em redes par-a-par

• Ano de defesa: 2006
• Autor(a) principal: Helio Marcos Paz de Almeida
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; UFMG
• Palavras-chave em Português: ataques Sybil; par-a-par; Arquitetura não-hierarquica (rede de computador); Computação móvel; Sistemas de comunicação móvel; Computação; Cliente/servidor (Computação); Redes de computação Protocolos; Sistemas de computação sem fio
• Link de acesso: http://hdl.handle.net/1843/SLBS-6Y5JAQ

Peer-to-Peer (P2P) systems use the cooperation between the nodes of the system to provide their services. However, it is possible for some participants to use resources provided by other nodes without giving anything back, a behavior known as free-riding. One way to force users not to act as free-riders is to use reputation mechanisms. In sucha mechanism, the behavior of nodes in respect to the provision of service to others is recorded and shared among the network nodes. With that information, a fair node can decide whether it should serve another node's request or not, based on the reputation of the requesting node. A previous dissertation (Bruno Gusmão Rocha, 2005) proposed a reputation mechanism based on game theory to exclude free-riders from a routing overlay network with good results. However, there are some problems that were not addressed by that work. One of those problems are Sybil attacks, which happen when a single user approaches thesystem with multiple identities, using them to get resources through mischief. Besides, that work required a global state in order to compute the participants reputation, what limits its scalability. Finally, the work was developed with a specic P2P environment in mind (routing overlay networks), but it would be interesting to see if the model could be applied successfully to other P2P systems, like those for le sharing. Our changes to the model focused those three problems. In a rst moment, we were able to exclude Sybil nodes with good performance, without hurting the service provided to fair nodes in the original model. We then altered the model to remove the need for a global state and observed that the new model stil presented performance close to the original model, even after that modication. Finally, we applied the reputationbased model to the BitTorrent le sharing network, achieving a better restriction to free-riders than that of the original network.