Impacto do comportamento dinâmico dos pares na eficácia de máquinas de busca par-a-par

Abstract

In an attempt to increase the spectrum of searchable information while attenuating scalability issues, Peer-to-Peer (P2P) networks have been viewed as an alternative way to design new Web search engines. However, the effectiveness of P2P Web searching may be severely limited by characteristics commonly observed in real P2P systems such as the dynamics of peer participation (churn). This study analyzes the impact of such issue on the effectiveness of P2P Web search engines. In order to estimate effectiveness boundaries, we focus our analysis on P2P network models with very high and low levels of peer knowledge about documents on the network. Our findings reveal that peer dynamic behavior could strongly affect search effectiveness even in optimistic scenarios: in networks where peers have a high knowledge about documents on the network, a significant fraction of queries suffer an impact on the quality of search of at least 26% still in highly stable scenarios. We also confirm that the impact of such issue in networks where peers have a lower level of knowledge can be even more intense (75%). We also evaluate content replication as a possible way to attenuate the effects of peer dynamic behavior on the effectiveness of P2P search engines. To this end, we analyze the effect of users downloading some ranked Web pages and adding them to their local collection. We observe that such strategy can significantly improve the effectiveness of P2P Web searching. In fact, the quality of results of networks with a very low level of peer knowledge about documents on the network can be significantly improved even in lowly stable scenarios. We then discuss some imposed challenges for the adoption of such solution. Actually, together with the high autonomy of peers and the absence of file-sharing benefits in replicating documents into the network, effectiveness of P2P Web search engines may strongly depend on new, application-specific incentive mechanisms for the users.