Managing peer database transactions

点击翻译按钮获取中文摘要

Many of the data sources upon which today's education, science, business, and government depend have been generating increasingly heterogeneous, networked, and vast amounts of data stored in a variety of formats. Our past work has contributed to building networks of relational databases (called peer databases, or simply peers) that make such a gigantic amount of data available for sharing. Peer databases share data using a peer-to-peer (P2P) network. A P2P network is point-to-point communication infrastructure which is used by peers to establish acquaintances prior to sharing data. Ensuring data sharing across heterogeneous boundaries in a semantically meaningful way has been at the core of our research. In collaroration with others, we have produced a prototype peer database management system (PDBMS) that uses the so-called mapping tables to solve the heterogeneity problem. Mapping tables are correspondences between data values as well as between schema elements at different peers. Using mapping tables, a PDBMS typically establishes acquaintances with remote peers, answers queries using data originating in acquainted peers, and propagate local updates to remote peers. Our research program has two main objectives which are to design advanced transaction models for the PDBMS setting, and to implement these transaction models on top our existing PDBMS infrastructure. Since PDBMSs are similar to traditional multidatabase systems in which the local data sources are not integrated, but rather coordinated, we closely examine the transaction models proposed for traditional multidatabase systems and relax many of their assumptions, as well as we add new ones that are more suitable for the PDBMS setting. The main issues we deal with are the relationship among local peer transactions, the consistency of the latter, concurrency control and recovery across a large amount of peers. We aim at making the vision of end-user oriented Internet data sharing possible. Users will only need to focus on their local data while the intricacies of the interoperability with remote peers will be handled as automatically as possible at the system level.