|Home > Publications > PhD theses > Numerical Analysis and Applied Mathematics (TW)|
A component environment of high performance scientific computing - Design and implementation
Advisor(s): Stefan Vandewalle and Herman Deconinck (von Karman Institute)
This dissertation focuses on the design and implementation of a framework for high-performance scientific and engineering computing. The framework creates an environment of modular components. These components can be reused and assembled together to form dedicated simulation tools to solve specific problems governed by partial differential equations.
This research is part of a long term investment by the von Karman Institute and many university partners, in the development of a platform for the solution of multi-physics systems using multiple numerical methods. The platform, named COOLFluiD, provides the environment where these different numerical techniques can coexist and work together. The present work is the design and development of the platform core, that allows such cooperation.
The main aim is the maximisation of three software quality objectives: efficiency, exibility and usability. These objectives are usually considered to be somewhat in conflict with each other. This thesis tries to show that they can be independently and simultaneously achieved within a single framework, provided certain techniques are employed. This research concentrates on the development of these techniques. Three sets of techniques are presented, and each set separates the software issues in a diㄦent direction. To derive these techniques the research followed certain keylines: first, it designed a framework architecture where components are reused in different applications; second, it focused on computational complex problems that demand high performance computing; and, finally, it defined a platform which is easily usable and developer friendly, where non-expert programmers can implement and research new numerical algorithms.
An assessment and validation of the three software qualities is provided to support the thesis and multiple applications are shown to demonstrate the capabilities of the framework.lirias 233965 / Doctadmin 3E080490 / text.pdf (8.9M) / mailto: twr team