Lehrgebiet Informationssysteme
FB Informatik
|
|
Lehrgebiet InformationssystemeFB Informatik |
|
|
Engineering Applications - a Challenge for the Next Generation of DBMS
Theo HärderUniversity of KaiserslauternP.O. Box 3049, 67653 Kaiserslautern, Germany e-mail: haerder@informatik.uni-kl.de
Full paper (postscript version compressed by gzip)Abstract:Today's DBMS are unable to meet the increasing requirements of emerging applications that would like to use a DBMS. To improve this situation a new generation of DBMS architectures adjusted to the de mands of such enhanced applications have to be developed. For this purpose, we have analysed the data management needs of a spectrum consisting mainly of engineering applications such as 3D-CAD, VLSI design, management of geographic information as well as diagnostic expert systems and have en countered both a modeling and a processing problem in today's DBMS. The lack of object orientation in current data models results in an inadequate mapping of application objects onto the data constructs pro vided by these data models; such burdensome object representations prohibit, in turn, effective manip ulation of application objects in a natural and smooth way. As a consequence, the flexibility and expres siveness of data models as well as the handling of application objects must be greatly improved before interactive design work can be supported.We try to explain which data model concepts and architectural decisions are necessary and why they can succeed in providing the anticipated data management support. In this framework, we discuss struc tural and behavioral object orientation, disjoint vs non-disjoint and recursive vs non-recursive object rep resentation, static vs dynamic object definition as well as the desired flexibility gained by symmetry of object access and dynamic object views. The most important architectural requirements consist of mul tiple application support, dynamic extensibility, efficient object manipulation, use of workstations, efficient remote data supply and failure isolation. Our design considerations have led us to a new kind of DBMS architecture which should better support the demands of non-standard applications. Such an NDBS ar chitecture consists of an application-independent kernel and an application layer providing application- specific operations. In a workstation/server environment the kernel is allocated to a server complex (po tentially consisting of multiple processors) whereas each application layer together with the correspond ing application is assigned to a separate workstation. The application layer plays a critical role for the performance of interactive design applications. For this purpose, main memory structures are maintained for all object representations 'under design' in a so-called object buffer. Another important aspect of the processing model is the transaction concept which has to enable local object handling by the application layer. By using a checkout/checkin mechanism we can obtain a perfect 'near by the application' locality of object processing which is considered the key to performance. Although the focal point of the paper is a tutorial-like description of application requirements and their consequences on DBMS design, we illustrate the recommended concepts by a practical data model and a DBMS architecture. The data model to be discussed is the molecule-atom data model (MAD model); the specific NDBS architecture is called PRIMA (prototype implementation of the MAD model). The MAD model as well as the PRIMA system were designed and implemented at the University of Kaiserslautern.
|