CSG Parallelism and Performance
and RWTH Aachen University
The CSG Parallelism and Performance supports performance engineering, in particular performance management, modeling, optimization, and parallel programming.
In the foreseeable future, the HPC community is approaching a new era of supercomputing known as exascale, which stands for the ability to execute 1018 floating operations per second. To reach exascale, the design of supercomputers becomes more and more complex and, hence, does the programming. For instance, hardware architectures are increasingly heterogeneous, forcing HPC applications to exploit different levels of parallelism. Ensuring both correctness and performance of applications running on such systems is a non-trivial task, requiring substantial expertise in HPC.
In the CSG Parallelism and Performance, HPC experts from RWTH Aachen University and TU Darmstadt, with Jülich Supercomputing Centre as an associated partner, bundle their expertise to foster the efficient and productive use of HPC current as well as emerging infrastructures by users from a broad spectrum of scientific disciplines. To achieve this goal, we provide user support, offer training, and organize workshops, targeting novice as well as advanced HPC users. In addition, we conduct research related to our service portfolio and, in particular, develop suitable tools.
- Parallelization support
- Correctness analysis and debugging currency errors
- Performance analysis and optimization
- Porting to new architectures / improving (performance) portability
- Enabling new HPC use cases
- Parallelism discovery
- Data race and deadlock detection
- Performance modeling
- Efficient coupling of AI with classic numerical simulation
- Archer (correctness checking for multithreaded programs)
- DiscoPoP (parallelization assistance)
- Extra-P (performance modeling)
- MUST (correctness checking for MPI programs)
Recent training activities:
- -41st VI-HPS Tuning Workshop (JSC/RWTH Aachen, Germany)
- PPCES / aiXcelerate (RWTH Aachen)
- ISC/SC Tutorials (PP101, OpenMP, Correctness, Performance)
- HiPerCH (HKHLR)
- RACE Tutorials (OpenMP / MPI)