Parallel and Distributed Implementation of Large Scale Industrial Applications

In many branches of industry, computer simulation has become an indispensable design tool that replaces a great deal of experiments in product design and thus reduces costs and time to market. However, simulation requires large amounts of computational resources. Parallel and distributed computing is a way to provide the required compute power especially to small and medium size companies or research facilities who cannot afford access to supercomputers. Interdisciplinary collaboration is required to port large scale industrial applications to parallel and distributed platforms efficiently. We report on results achieved in such a joint project of two medium size companies and two academic research groups. In a well-defined software engineering process, we have parallelized a state of the art CFD simulation package for execution on a wide range of hardware platforms based on PVM and MPI. Test runs show very good performance and scalability. We have worked out an object oriented redesign of the CFD code and implemented it in C++ and Java. To optimize the use of workstation clusters, we have developed a resource management system that dynamically allocates idle resources to parallel batch jobs. We also address the problem of result visualization which is about to become a bottleneck if computation executes efficiently in parallel.