Abstract:
Peer-to-Peer (P2P) data integration systems have recently attracted significant attention for their ability to manage and share data dispersed over different peer sources. While integrating data for answering user queries, it often happens that inconsistencies arise, because some integrity constraints specified on peers' global schemas may be violated. In these cases, we may give semantics to the inconsistent system by suitably " repairing " the retrieved data, as typically done in the context of traditional data integration systems. However , some specific features of P2P systems, such as peer autonomy and peer preferences (e.g., different source trusting), should be properly addressed to make the whole approach effective. In this paper, we face these issues that were only marginally considered in the literature. We first present a formal framework for reasoning about autonomous peers that exploit individual preference criteria in repairing the data. The idea is that queries should be answered over the best possible database repairs with respect to the preferences of all peers, i.e., the states on which they are able to find an agreement. Then, we investigate the computational complexity of dealing with peer agreements and of answering queries in P2P data integration systems. It turns out that considering peer preferences makes these problems only mildly harder than in traditional data integration systems.
