Database consistency is the normal measure by which databases are determined to be correct. The normal technique for ensuring consistency is to implement concurrency control techniques that ensure serializability. However, ensuring serializability in a multidatabase system (MDBS) can be not only difficult, but expensive. This paper presents a technique that can be used to ensure data consistency in an MDBS without requiring global serializability. Instead, a weaker restriction, called two-level serializability (2LSR), is enforced. The benefit of this approach is that 2LSR is much easier to implement and ensure in an MDBS system than regular serializability.
I found the work to be technically accurate, but at times difficult to read. The presentation and organization are fine, but there is a lot of notation, some of which could have been avoided. As it is, the paper’s audience is probably limited to theoretical database researchers specializing in consistency issues for distributed and MDBS systems. Most of the notation is important to a thorough understanding of the theory, but is not needed to comprehend the overall ideas and approaches provided in the paper.
The paper presents some very interesting ideas that will interest database implementers and programmers. Thus, the results are of practical interest. In section 5, five different approaches to implementing 2LSR in an MDBS are presented. A nontheoretical database programmer would find that these approaches present reasonable techniques that can be used to ensure consistency. Not all of them are applicable in all situations, but each indicates what restrictions on transactions must be made to ensure that a 2LSR execution will guarantee consistency.
This paper not only provides an excellent theoretical overview of the consistency issues related to MDBS systems, but proposes practical approaches that can be implemented to ensure consistency without enforcing global serializability. However, it is not for everyone. The technical content requires not only a database but also a mathematical maturity not typically possessed by the average programmer. Thus I recommend this paper only to researchers in the area.