Transaction scheduling in dynamic composite multidatase systems

Abstract

Multidatabase systems based on a single monolithic multidatabase server are not realistic. Their performance and administration do not scale with increases in the radius of service or the number of component databases under their control. We propose that a composite multidatabase architecture that consists of multiple, possibly heterogeneous, peer multidatabase servers distributed on a communications network is inevitable. Global transactions should be able to span multiple multidatabase servers, sometimes forcing multidatabase servers to act as component database systems. Particular focus is given to the problem of guaranteeing the correct execution of interleaving global transactions across multiple multidatabase systems. Correctness is based on global serializability. Three algorithms for maintaining global serializability through transaction ordering during dynamic multidatabase composition are proposed. We examine the restrictions of these algorithms and the scalability of their performance.