Projects in Multidatabases

Principal Investigator: P.K. Chrysanthis
Sources of Funding: National Science Foundation

This project examined atomic commitment in MDBSs where each local database externalizes an atomic commit protocol. Specifically, we investigated the issue of compatibility among two-phase variants and showed that supporting a visible prepare to commit state is not enough for their successful integration in an operational fashion because the outcome of some transactions might have to be remembered forever.  We defined an operational correctness criterion that allows terminated transactions to be forgotten. Based on the proposed operational correctness criterion, we developed a two-phase commit (2PC) protocol, called Presumed Any, that integrates the presumed nothing, presumed abort and presumed commit 2PC variants despite their conflicting presumptions about the outcome of transactions and without violating the autonomy of the local database systems.

Furthermore, based on the same principle, we have proposed three strategies that allow a multidatabase system to dynamically adapt to the most appropriate 2PC variant at any given time. This is achieved by structuring the agents so as to act as participants in the case that the selected commit protocol is different than the one of their corresponding local database system, while at the same time, they act as coordinators for their corresponding local database system. Depending on the strategy, agents may switch protocols on a per transaction, on a per participant and on a per coordinator basis.

Consistency in Multidatabase Systems

Principal Investigator: P.K. Chrysanthis
Sources of Funding: National Science Foundation

The use of many diverse database management systems (DBMSs) in an organization stems from reasons concerning suitability, ownership and security. However, in these organizations, new applications, for example, design or decision-making environments, require information stored in several of these pre-existing databases. Multidatabase systems (MDBSs) respond to the need of organizations by supporting access to data stored in multiple databases managed by autonomous and possibly heterogeneous database systems. Because of the autonomy and heterogeneity of the component database systems, the issue of data consistency is fundamentally different between distributed DBMSs and MDBSs. Our goal in this project is to understand the significance in practical terms of the correctness and consistency criteria in MDBSs by treating the issue of consistency formally. In particular, we desire to apply a uniform specification technique to express the various correctness and consistency criteria that have been proposed. A formal specification technique will facilitate the proposal of new correctness and consistency criteria in a precise manner as well as help in the design of the appropriate algorithms for maintaining consistency.