Go to Paper
Return to GeoComputation 99 Index
Towards a Framework for High-Performance Geocomputation: Handling Vector-Topology within a Distributed Service Environment
DOWERS, Steve (firstname.lastname@example.org), GITTINGS, Bruce and MINETER, Mike, University of Edinburgh, Department of Geography, Drummond Street, Edinburgh, EH8 9XP, U.K.
Key Words: high-performance computing, vector topology, OpenGIS, software libraries, service architectures, algorithms
High performance computing has been shown to be a critical technology to realise current and future spatial data handling requirements. While commercial parallel processing approaches within DBMS are providing the performance for underlying integrated data models, the acceleration of spatial analyses remains to be addressed. This paper investigates the scope for a framework that integrates parallel implementation within a service architecture such as that proposed by the Open GIS Consortium.
The application of parallel processing to spatial analyses requires flexible strategies for decomposition of data if a scalable solution is to be achieved. Such strategies are relatively well-known for grid, raster, and digital terrain data, and for the spatial analysis of point data. These techniques are not established for vector-topological data. Furthermore, the application of parallel processing to geocomputational algorithms has been limited to specialists and has not made the transition to the operational area of the end-user as has happened with database technology. Previous work by the authors has explored the issues involved in designing a generic approach to the handling of vector topological data in parallel and to the development of parallel libraries that encapsulate the complexity of the parallel implementation. The goal has been to allow the algorithm implementor to concentrate on the specifics of the modelling process, and also take advantage of the performance gains from parallel architectures in a portable manner.
Some success in this approach has already been reported in previous publications, while current projects are testing implementations of potential components of a parallel library concerned with input, distribution, collation, and output. Central to the details of the these implementations is the choice of data models and the representations of features, geometry, and the topological relations between them. Existing work by the authors has been based on the use of Neutral Transfer Format (NTF) Level 4 model which is part of a British Standard. Current work is exploring the issues of the interface between parallel applications and other systems and services. These systems may include thin clients, spatial analysis engines, and databases with spatial capabilities. The OpenGIS Abstract Specification Model addresses many of the issues associated with these interfaces, including data models and data exchange. This paper will explore the implications of this and related models for our existing design work, and will identify the role of parallel libraries and high-performance applications in the general framework of a service architecture. Debates are ongoing as to the benefit of stored versus generated topology with integrated spatial data stores; we shall explore the generic nature of algorithm designs and the extent to which they can apply in these differing contexts.
British Standards Institution (1992) Electronic Transfer of Geographic Information (NTF, BS 7567, British Standards Institution, London, U.K.
Healey, R., S. Dowers, B. Gittings, and M. Mineter, (1998) Parallel Processing Algorithms for GIS, Taylor & Francis, London, U.K.
Mineter, M.J. and S. Dowers (in press) Parallel Processing for Geographical Applications: A Layered Approach, Journal of Geographic Systems.
Open GIS Consortium (1998) The OpenGIS Abstract Specification Model, http://www.opengis.org/techno/specs.htm