Parallelization of large scale Industrial Aerodynamic applications on the IBM RS/6000 SP

Abstract

In the last years we have experienced a dramatic change in scientific supercomputing, shifting from the parallel vector processors to the RISC based multiprocessor systems. A brief look into the TOP500 list [3] shows that nowadays the RISC based multiprocessor systems have already gained a majority regarding both the number of installations and the Megaflops installed. Even when considering the cost-effectiveness for large commercial application programs these multiprocessor systems have taken the lead now [13]. Due to this fact a change of the programming paradigm for the application software will be the consequence. If scalability is the crucial issue only distributed memory multiprocessor systems offer the breakthrough, whereas shared memory multiprocessor systems cannot keep pace. If, however, speed is the crucial issue only explicit message passing code guarantees high efficiency, whereas at the present time High Performance Fortran (HPF) code does not [5]. Therefore the most promising high performance computing approach to industrial applications are distributed memory multiprocessor systems based on the programming paradigm of explicit message passing. In 1993 a joint national project POPINDA (POrtable Parallelization of INDustrial Aerodynamic applications) funded by the German Ministry of Education and Research (BMBF) was launched with the aim to develop two portable parallel block structured CFD codes for complex flows around aerodynamic con-