Realizing the benefits of high performance computing for solving environmental and other problems: what do we need beyond faster hardware?

Abstract

A common viewpoint is that high performance computing offers the potential to revolutionize our approach to modern scientific and engineering practice. Unfortunately, in the debate about just how to realize the benefits much of the discussion is narrowly focused on just the hardware. What is missing is a more pragmatic assessment of how to reduce the elapsed time between problem definition and its solution. This presentation will discuss the need for a seamless integration of such elements as algorithm development, problem mapping, high level compiler technologies, documentation and display of results. Particular attention will be given to the issue of how to capture the potential offered by massively parallel processing and Gbit/s network based supercomputing. The talk will be illustrated with many practical examples that range from a combinatorial optimization based approach to the design of new chemical plants, computational chemistry, resource scheduling in the presence of uncertainty and scientific visualization. Several new algorithms will be presented for the solution of large systems of differential equations. The new methods, which require modern compiler technology to ensure optimal problem mapping strategies, are 3-5 orders of magnitude faster than existing techniques. The final component of the presentation will be a discussion of the educational issues and in particular the :need for more programs in scientific and engineering computation.