Maintaining multidatabase consistency: The quasi-serializability approach

Weimin Du, Purdue University

Abstract

This thesis studies a new approach to maintain multidatabase consistency, the quasi serializability approach. Quasi serializability is a correctness criterion for multidatabase consistency. It is weaker than serializability in that it controls interactions between global transactions only. Quasi serializability is easier to ensure when local database systems are autonomous and is suited for multidatabase applications where local transaction management systems maintain serializability of local executions. Quasi serializable executions maintain multidatabase consistency if local transactions at different sites do not affect each other. Based on the quasi serializability theory, mechanisms that maintain multidatabase consistency are presented. For non-replicated multidatabase systems, an access graph algorithm is proposed to ensure quasi serializability, along with a technique to prevent undesirable remote interactions between local transactions at different sites. For replicated multidatabase systems, a decentralized strategy for replication control is proposed. The mechanisms are different from the existing ones (based on serializability) in that they do not violate local autonomy, provide a higher degree of concurrency, and abort fewer global transactions. In combination, the two algorithms we proposed demonstrate the viability and desirability of maintaining multidatabase consistency in quasi serializability approach, rather than in the conventional serializability approach.

Degree

Ph.D.

Advisors

Elmagarmid, Purdue University.

Subject Area

Computer science

COinS