Supporting Autonomy for Information Systems in a Changing Environment

In the development of information systems, object-oriented speciication (e.g. Troll 4]) is useful for conceptually modeling of the universe of discourse. Viewing an information system as a collection of communicating objects is close to the intuitive perception of such systems on a conceptual level. Nowadays complex information systems (e.g. knowledge bases) have to be exible to accommodate for changes in organization structures. Mandatory features are distribution for decentralization support and scalability for modular system increase. Furtheron processing in parallel enables an important speedup. In general, distributed database systems 7, 6] have to deal with a lot of tradeoos, e.g. data replication versus data transfer, reuse versus autonomy, transparency versus eeciency. Our goals in gaining distribution for an object-oriented data model are: (1) Execution autonomy: For decentralization support, federation does neither interfere with local operations nor is any knowledge of the federation needed for performing local operations. (2) Transparency aspects: For scalabil-ity support transparency of location and migration enables objects to be used without knowledge of their location and movement of objects within a system without aaecting the operations. In conjunction with horizontal and vertical class partitioning, the system could be expanded in scale without changing the spe-ciication or the references. (3) Access parallelism: For speedup support, access time to partitioned classes has to be decreased by parallelized mechanisms. To gain this kind of autonomy, transparency and parallelism for object-oriented speciication, at least the following problems have to be achieved: Object-oriented speciication consists of a global abstract description as conceptual model without non-functional requirements (e.g. distribution, persistence, exception handling). A distributed infrastructure consists of a set of interconnected loosely coupled nodes with own processors and disk-spaces. On whichìevel' distribution has to be combined with the conceptual model? A complex information system structure, which is described as conceptual entirety, is transparently distributed over several nodes. Important is a partial use of the information system , although not all nodes are available or reachable due to a side failure. How to achieve a maximum of node autonomy for a distributed system (C. J. Dates rst (of 12) rule for distributed database systems 2]) supported by the distributed schema catalog? Existing approaches in the area of database systems mostly do not pay enough attention to the opportunities of autonomy and scalability, which are getting increasingly important by a new generation of parallel hardware clusters 3].