A retrospective of the Darmstadt database system project, also known as DASDBS, is presented. The project is aimed at providing data management support for advanced applications, such as geo-scientific information systems and office automation. Similar to the dichotomy of RSS and RDS in System R, a layered architectural approach was pursued: a storage management kernel serves as the lowest common denominator of the requirements of the various applications classes, and a family of application-oriented front-ends provides semantically richer functions on top of the kernel. The lessons that were learned from building the DASDBS system are discussed. Particular emphasis is placed on the following issues: the role of nested relations, the experiences with using object buffers for coupling the system with the programming-language environment and the learning process in implementing multilevel transactions.>

The multi-layered architecture of the DArmStadt Data Base System (DASDBS) for advanced applications is introduced DASDBS is conceived as a family of application-specific database systems on top of a common database kernel system. The main design problem considered here is, What features are common enough to be integrated into the kernel and what features are rather application-specific? Kernel features must be simple enough to be efficiently implemented and to serve a broad class of clients, yet powerful enough to form a convenient basis for application-oriented layers. Our kernel provides mechanisms to efficiently store hierarchically structured complex objects, and offers operations which are set-oriented and can be processed in a single scan through the objects. To achieve high concurrency in a layered system, a multi-level transaction methodology is applied. First experiences with our current implementation and some lessons we have learned from it are reported.