Distributed parallelism: impacts on GIS and collaborative spatial decision-making

Because GIS software is used most effectively to address public policy problems if several alternatives, each representing the interests of different stake-holders, can be generated and evaluated during a meeting, the response time of GIS-based models can assume considerable practical importance. In this paper we present results from research that evaluates a new approach to the analysis of computationally-complex GIS-based models, with the aim of improving the performance of networked, collaborative decision-making environments. The first element of the paper centers on the use of a network of workstations (NOW) to support parallel interpolation for a large problem (1024 x 1024 cells with 10,000 control points). Computational experiments using this approach achieve a level of performance that compares favorably with a dedicated parallel supercomputer (Cray T3D). The experiments also show general promise for a large class of GIS algorithms that can be divided into large sub-problems. These results are especially significant since networked workstations present a ready source of computing power that can be accessed and used during the period of a typical group meeting.