Project Summary

The PINEAPL project will be a coordinated effort to produce a general purpose library of parallel numerical software suitable for a wide range of computationally intensive industrial applications and to port several application codes which use this library to parallel computers; the initial coverage of the library will be determined by the requirements of the applications. The project includes a number of key features:

The end-users in the consortium will take advantage of the opportunity provided by the project to introduce new mathematical models and techniques into existing applications. Reduced wall-clock execution times will allow users to experiment with more configurations in their simulations, thereby achieving better designs and gaining new physical insights into their problem. Larger problems may be solved or greater accuracy in the solution may be achieved with these new techniques. The consortium will use existing tools, which offer information about mathematical models and numerical methods to build confidence in the new techniques.

The major result from the project will be a portable library of high performance mathematical software suitable for both the application codes and for other aspects of parallel computing such as optimizing load balancing and communication. There are more than 3000 industrial and academic organisations using general purpose numerical libraries in Europe alone. The demand for similar libraries for distributed memory parallel machines is increasing as more of these users take advantage of networks and clusters of workstations and distributed memory computers. The relevance of the library software to industry will be demonstrated in the project applications and further assured by advice received from the other industrial interests.

Target machines for the project will range from low cost networks of workstations that are common in SMEs to high-end distributed memory parallel computers. The objective in producing a numerical library is to encapsulate efficient computation (and communication) into easily used modules, insulating the user from the details of these complex areas of expertise.