Towards Peer-to-Peer Information Management Systems

Peer-to-Peer (P2P) is a fully distributed communication model in which parties (aka peers) have equivalent functional capabilities in providing each other with data and services. Because they are totally autonomous, peers are independent of other peers in which data they store, in which other peers they interoperate with, and in how they process peers’ service requests. While the cost of participation for a single peer is limited to seldom answering of peers’ requests, the benefit of a peer grows rapidly as more peers take part. In order to reach this effect, peers coordinate the processing of the requests by transitively passing data to one another. The benefits of using P2P as the underlying architecture for building e-commerce and Semantic Web applications has been realized. However, there are still some essential challenges to be addressed. We summarize them in three groups: • the decentralized open-ended nature of P2P makes it very hard to locate peers which have necessary data, as well as to control the scope of peers’ interaction; • the autonomy of peers implies that no global agreement on data representation languages, on the contents and the location of the data can be made; and, • partly as a consequence of the first two items, the quality of answers in a P2P network is unpredictable, and it cannot be measured by standard data quality metrics such as correctness and completeness. The present thesis is one of the few first dissertation works in the area of P2P information management systems. It provides a detailed account of what P2P information system is and how it relates to the previously developed systems for data and information management. Particularly, we consider conventional databaseand ontology-based information systems and show in what respect they are different from P2P ones. We also show how application space is divided among the conventional and P2P information systems. Apart from this, we propose a novel information management model, which addresses the first two challenges described above. To address the third challenge we introduce the notion of good-enough answers and show in what respect it is different from the standard quality metrics, well studied in the data quality literature. Finally, based on the proposed information management model, we developed a novel query answering algorithm, which supports cyclic topologies and a dynamic system behavior.