Bundling: A new Construction Paradigm for Persistent Systems ?

Over the past years, successful research efforts have been devoted to implementation aspects of database management systems. This research has mainly considered the implementation of single DBMS-components and techniques for selected DBMS-tasks (e.g., query optimization, transaction management). Nevertheless, the systematic construction of DBMSs as a whole has been considered only scarcely. Urgent problems such as domain-specific architecture models for DBMSs or DBMS-construction processes have received attention only recently. Such problems gain ever more importance as the trend is towards functionally more and more powerful persistent systems. Since it is apparent that all the requirements that might be posed to a DBMS can no longer be satisfied by a single, monolithic system, database technology must provide for a "construction theory" that allows to assemble adequate systems out of reusable components in a cost-effective way. Thus, our belief is that in the long term, DBMSs will be constructed by putting together components. This approach-called "rebundling- relies on reuse of adequate components, which are stored in repositories. The process that identifies and extracts these reusable components (out of virtual or real systems) is called "unbundling". In this paper, we first view the prerequisites and characteristics of unbundling and rebundling from a software engineering point of view. We then sketch how DBMSs can be constructed following the bundling paradigm, survey recent work in database technology related to un- and rebundling, and identify open problems. We then leverage unbundling to general heterogeneous, distributed persistent systems whose behavior is defined by processes. In this kind of system, bundability is an even more important requirement, because the components often are entire applications that cannot be rewritten. Using a scenario, we show how bundling can be applied for this kind of information systems.