The INSIDE project: Some remarks concerning Control Flow in ’Real Life’ Parallel Software and the Parallelization of the p-Version of FEM

This contribution describes intermediate results and motivates approaches for two topics relevant for the ESPRIT project INSIDE. After a short sketch of the project itself the paper first discusses the needs for a software tool to control informationand data flow in an HPC network to be used for real life applications and properties of the resulting prototype software. The second part of the paper motivates the use of the p-version of the FEM and adresses some reasons why this method is expected to be superior in parallel efficiency compared to standard h-version codes when targeting at structural analysis problems relevant to the INSIDE project. 1 The INSIDE project The INSIDE project (Integrated Simulation and design system for civil and structural engineering, ESPRIT project 20.216, [1]) was initiated in 1996 with the objective to develop an HPCN-based integrated CAD-FEA system for civil and structural engineering, supporting interactive optimization of constructions using distributed and parallel software on workstation clusters and low-cost parallel computers. The speedup of computation necessary for this interactive structural design will be achieved by combining methodological improvements (mesh adaptivity and p-elements) to reduce the required number of unknowns without loss of accuracy and parallel computing supported by high performance networking. A multitude of software modules interact to form the system: The user works on a PC using a CAD program to define a structural problem. A so-called monitor program allows (as an additional feature within the CAD surface) to define a virtual parallel machine consisting of UNIX-workstations and to start and monitor a distributed simulation which itself consists of several task specific groups of processes, each of which is organized according to a master-slave hierarchy.