Parallel programming, algorithms and methods
ZIH is actively contributing to research on parallel programming models for high performance computing (HPC) via standardization and implementation. This way it helps provide one of the basic prerequisites for the usage of supercomputers in science and engineering. At the same time, ZIH is working with users from various scientific application domains to design and implement parallel algorithms, in scalable ways with appropriate programming languages and parallelization models.
The Message Passing Interface (MPI) dominates parallel programming nowadays. Still, there is a range of established or novel parallelization models that can either be used in combination with MPI or as alternatives to MPI. One such group of models targets shared memory parallelism (SMP) using multi threading and tasking, e.g., OpenMP, OpenACC, and PThreads. Another promising group follows the partitioned global address space (PGAS) paradigm to combine the SMP approach with distributed parallelism, e.g., CoArray-Fortran (CoF), OpenShmem, and GASPI. Finally, there are special programming models leveraging the usage of accelerator devices, e.g., for GPGPUs (general purpose graphics processing units) with NVIDIA CUDA and OpenCL.
HPC offers enormous computing potential for a large number of disciplines from science and engineering. Yet, it can only be fully exploited when the knowledge from the application domains is combined with the best available algorithms, as well via matching parallelization and implementation methods. Towards this goal, ZIH works in close cooperation with application domain experts towards efficient and scalable application programs.
Current Key Topics and Projects
- DASH: Hierarchical Arrays for Efficient and Productive Data-Intensive Exascale Computing
- Standardization of OpenMP and OpenACC
- cfAED: Center For Advancing Electronics Dresden
-
CRESTA: Collaborative Research into Exascale Systemware, Tools and Applications
-
EXPLOIDS: Explicit Privacy-Preserving Host Intrusion Detection System
-
GASPI: Global Address Space Programming Interface
-
GlioMathDD: Identification of new biomarkers and therapeutical targets for secondary glioblastoma with the help of innovative methods from Systems Biology
- GCOE: Dresden GPU Center of Excellence
-
HPC-OM: HPC-OpenModelica für Multiskalen-Simulationen technischer Systeme
-
H4H: Optimierung von HPC-Anwendungen auf hybriden Architekturen
- HPC-Flis: HPC framework for the solution of inverse scattering problems on structured grids and its application for 3D imaging
- IPCC: Intel® Parallel Computing Center TU Dresden
- MEPHISTO - Metaprogramming for Heterogeneous Distributed Systems
- NGSgoesHPC: Scalable HPC solutions for efficient genome analysis
- PARADOM - Parallele Algorithmische Differentiation in OpenModelica für energietechnische Simulationen und Optimierungen
- Parallel coupling framework and advanced time integration methods for detailed cloud processes in atmospheric models
- ScaFES: Scalable Framework for Explicit Solvers
- SFB 609: Elektromagnetische Strömungsbeeinflussung in Metallurgie, Kristallzüchtung und Elektrochemie
- VEKTRA: Virtual Development of Ceramic and Composite Materials with Tailored Transport Properties