What is a Problem Solving Environment?

The standard method for running a numerical simulation of problems, including EHL, is to

1. Set input parameters in code
2. Compile code
3. Run code, saving results to disk
4. Post process results and load into visualisation package
5. Examine results to see what changes need to be made to the input parameters and return to (1)

Typically, one has two choices with using PSEs. Either build your own from scratch or use a framework which enables you to interface whichever parts you are developing yourself with standard tools available in your chosen package. For example, fast, memory efficient isosurfacing algorithms or complex OpenGL graphic packages need not be written by anyone only interested in solving a numerical problem.

Why do EHL in a PSE?

In EHL the PSE has several applications. First and foremost, the ability to visualise numerical results to enormous datasets is very important. The added insight that can be gained from a human brain interpreting a 3d coloured rendering can far exceed that possible from looking at a few key numbers of output results. For transient cases then being able to see an animating image can add even more insight.

Computational steering - the ability to change input parameters once the application has started running, means that cases can be run and the influence of individual parameters be analysed by making interactive changes. If a case is running and numerical convergence is poor, or solution resolution is too coarse then the grid refinement level used can be increased. Computational steering is particularly in collaborative sessions, described below.

Constructing the EHL PSE

IRIS Explorer is a commercial visualisation package. It operates by having a 'map' of connected 'modules' joined together by a 'dataflow pipeline'. Different modules perform operations on the data coming in, be it from previous modules in the map, or or widgets on the module interface. Output is either by sending data down the map to later modules, or through visualisation on the screen. In IRIS Explorer each module is a separate process on the computer. Communication between modules is through a shared memory arena, and there are various predefined datatypes covering everything from highly regular datasets, through unstructured grids to geometry.

Eclipse in IRIS Explorer

Ellipse in SCIRun

SCIRun was developed as a PSE for a very specific application. This has since been extended to a more general package and is now available for free, non-commercial download. It is similar in style to IRIS Explorer with 'modules' joining together to form a 'network'. The most major difference is that the entire of SCIRun is one large, multi-threaded process. Ellipse, the SCIRun PSE, can be seen running in this movie of progressive refinement of grids.

The EHL PSE has been constructed in each of these by embedding the numerical solver as one module. In the IRIS Explorer version the input variables are all available in separate modules feeding into the main solver, whereas SCIRun has all the available controls on the same module. Outputs are solution plots for the pressure and the surface geometry.

A paper comparing the differences in using the two systems was presented by myself at the ICCS 2002, published in Lecture Notes in Computer Science, vol 2329, pp 523--532.

Collaborative EHL PSE in IRIS Explorer

Using this environment between ourselves and Shell at Chester is one of the aims of the current e-Science Grid project linked below.

Further Links

• COVISA - Collaborative Visualization the PhD Thesis of Jason Wood.
• COVISA-g - an e-Science demonstator collaborative PSE on the Grid