Strategic research priorities
For RWTH Aachen University, the Computational Engineering Sciences (CES), consisting of engineering and material sciences as well as physics, chemistry and computer science, are key topics that are documented in special research areas and other interdisciplinary activities.
The HPC infrastructure and support projects at RWTH Aachen University will be bundled to combine their individual strengths and create an HPC ecosystem that combines best practices in HPC and research data management. This comprehensive and powerful ecosystem has both the breadth and depth to support the development of scalable algorithms and software for the study, design, construction, evaluation and production of engineering applications. These applications are critical to technological advances in business and society.
Application and Methodology Combined: SDLs and CSGs
Since the foundation of NHR4CES, application-oriented HPC research has been carried out in the Simulation and Data Labs (SDLs) and the Cross-Sectional Groups (CSGs). Here, RWTH uses the expertise of the scientists at RWTH to drive the SDLs forward in terms of cutting-edge research and feed back into our degree programmes and international standardization efforts.
A large part of the application-oriented research is aimed at sustainably improving the efficiency of high-performance computers and their applications. The results achieved are used directly, for example in service units of the Cross-Sectional Group Parallelism and Performance.
HPC research on parallel programming and performance tools is carried out at the Chair of High Performance Computing (HPC) / Computer Science 12 in the Department of Computer Science, which is closely linked to the IT Center. The HPC group's research includes performance and correctness analysis of parallel programs as well as research into energy efficiency, programmability and total cost of ownership.
Standardization and OpenMP Development
In the context of parallel programming models, for example, the focus is on the further development of the OpenMP industry standard. This industry standard for parallel programming is used in countless applications to make the performance of modern multi- and many-core systems and accelerators accessible. Since 2006, RWTH Aachen University has been a member of the OpenMP Language Committee, which continuously develops the standard, as well as the OpenMP Architecture Review Board, the body to which this standard belongs. RWTH Aachen University's major contributions to OpenMP include the topics of thread affinity, memory management, improvements in tasking and the OMPT tool interface.
In the correctness analysis of parallel programs, the MUST tool for the automatic error analysis of MPI programs is developed and the development of ThreadSanitizer for the correctness analysis of thread-parallel programs is improved. Parallel programs are also examined for possible errors such as race conditions and deadlock or incorrect use of the standard and the results are presented to the user in a clear manner. It is also possible to analyze hybrid-parallel programs.
Furthermore, approaches for evaluating and improving productivity in the use and programming of HPC systems are examined. In particular, the aspect of overall cost optimization is considered and the benchmarks of the Standard Performance Evaluation Corporation (SPEC) are further developed.
This research is mainly supported by third-party funding from the EU, the BMBF and the DFG. The results are continuously used to improve the operation and efficiency of the HPC systems at the IT Center of RWTH Aachen University.