The Visualisation of Large Unstructured Grid Data Sets

Numerical Grid Generation in Computational Field Simulations, Ed. M. Cross., B. K. Soni, J. F. Thompson, J. Hauser, P. R. Eiseman, Proceedings of the 6th International Conference, held at the University of Greenwich, pp.899-909, July 1998

MESHING
RESEARCH
CORNER

Department of Civil Engineering
University of Wales Swansea
Singleton Park, Swansea, 5A2 8PP, U.K.

Introduction
Despite the significant advances in the computational and rendering capabilities of modern high performance graphics workstations, there remain visualisation applications which require the computational power available only to todays parallel and vector supercomputers.

This is certainly true in the field of computational engineering, in particular, OFD (Computation Fluid Dynamics) and OEM (Computational Electro-Magnetics), where generating and performing calculations on meshes of the order of 10 million elements are becoming routine in the research environment, and calculations on meshes of the order of 50 or 100 million elements are becoming possible [3]. Producing interactive displays of meshes, and their associated data, of this size is beyond the capabilities of modem graphics workstations. In order to solve this problem, distributed processing needs to be employed using a combination of graphics workstations and supercomputers under the close control of a single application.

The supercomputer is able to supply the computational power, memory storage and I/O transfer rate needed to perform the number-crunching calculations on the large data sets and the graphics workstation can transform the results into high-resolution, colour images with the ability of allowing the user to interact with the models in real-time.

The grid generation phase of the computational cycle is where rapid visualisation of data is first required. It is straight-forward to quantify where there is a need if grid generation is thought of as three phases; Specification, Generation and Evaluation.

The Specification phase is where the user interacts with a geometry and specifies all the required input data, such as grid point density parameters. The Generation phase is where, given all the required input parameters, the grid is generated using relatively compute-intensive algorithms. The Evaluation phase involves the user evaluating the suitability and quality of the generated grid. Recently, the Specification and Evaluation phases utilise advanced computer graphics workstations and algorithms embedded within an easy-to-use graphical environment.

It is often the case that with present day grid generation algorithms, the time required for the Generation phase is very small compared with the time spent in the Specification and Evaluation phases. Typically, using a Delaunay grid generator, an unstructured grid of 2 million elements will take less than 30 minutes to generate on a modest workstation [1,2].

Therefore, the key to efficient problem turn-around, is effective interaction in the Specification and Evaluation phases. For small data sets, it is easy and efficient. . .