Archive

In conjunction with the introduction of Claix-2018 a joint technical collaboration was established between RWTH and NEC to evaluate porting and the performance of different production codes to the vector architecture SX-Aurora Tsubasa. In this context investigations of XNS, ZFS, VASP, CIAO, Quantum ESPRESSO and CP2K were realized. Furthermore an OpenMP Device Offloading Infrastructure was developed and evaluated with the SPEC-hpc and SPEC-Accel Benchmarks. This effort was complemented with a dissemination of this knowledge at different Workshops.

Project duration: January 2017 - June 2020
Funding: DFG Performance Engineering for Scientific Software

The aim of the ProPE project was to develop a blueprint for a sustainable, structured and process-oriented service infrastructure in the field of performance engineering (PE) of application codes in high-performance computing with a focus on German Tier-2 and Tier-3 HPC centers. This should enable application scientists to use HPC systems with demonstrably optimal hardware resource utilization and thus reduce IT-related costs. To this end, the HPC centers in Aachen, Erlangen and Dresden have jointly evaluated, developed, and implemented processes, methods, and tools.

The project and the Pro-PE initiatives coordinated by Aachen were reported on several times in the IT Center Blog: On the one hand, information was provided about the supra-regional support for HPC users, which enables enquiries from HPC users to be answered in a structured manner and according to expertise across all centers. Furthermore, the idea and implementation of the HPC- WIKI created in ProPE was explained. In addition, an HPC performance monitoring system was set up at all centers as part of the project. The automatic monitoring at the IT Center shows, for example, the flop/s and memory bandwidth of a computing node and thus enables a low-threshold analysis of the performance data of all HPC jobs. Since August 2023, this data has not only been available to the HPC experts at the IT Center but can also be viewed by all HPC users themselves. Information on this can be found on IT Center Help.

The experiences from the ProPE project were published in a white paper, which also serves as a blueprint for the German Council of Science and Humanities for nationwide use. The IT Center also reported on the project results and their further applicability in its 2021-22 annual report. (only in german)

The time-dependent behavior of parallel simulation codes -- in particular when adaptive algorithms are employed -- is often irregular, making the understanding of performance dynamics an essential prerequisite for program optimization. While existing performance analysis tools typically provide detailed information along spatial dimensions like processes and nodes, performance dynamics has so far been a neglected aspect.

To support developers of parallel simulation codes with optimizing their codes, the LMAC project aims to extend the established performance analysis tools Vampir, Scalasca and Periscope with new functionality to automatically examine performance dynamics. In addition, the University of Oregon, associated partner, complements the project with corresponding extensions to the performance tool TAU. Further information on the project can be found on the VI-HPS website.

SILC - Scalable Infrastructure for the Automated Performance Analysis Nowadays applications in the field of high performance computing have to utilize hundreds or thousands of compute cores to make efficient use of modern compute architectures. A variety of performance analysis tools can be used to assist programmers in the process of application tuning to utilize the cores in a good way. These different analysis tools all collect performance data which is then analyzed and presented to the programmer in a different way. For a programmer this means that similar data is collected with different tools using different options. Within the SILC project a uniform measurement infrastructure (Score-P) was developed to support multiple different performance tools. This eliminates the need to measure performance data several times with different tools, but it still allows using all different analysis tools to interpret the data. Tools which can use the data generated by Score-P are Vampir , Scalasca, Periscope and TAU.

Partners of the SILC project, which was funded by the German Ministry for Education and Research (BMBF), are the RWTH Aachen University, TU Dresden, TU Munich, GNS Gesellschaft für numerische Simmulation mbH, Forschungszentrum Jülich and the associated partners University of Oregon and German Research School for Simulation Science. Further information on the project can be found on the VI-HPS website.

The mechanical behavior of semi-crystalline thermoplastic parts is greatly influenced by the microstructure. Therefore the accurate prediction of microstructure is important for part design. Today, simulation programs like the microstructure simulation software SphäroSim which has been developed at the RWTH Institute of Plastics Processing (IKV) restrict the simulation to small areas of the part. In order to develop algorithms that can predict finer structures and to simulate the microstructure in an entire injection moulded part, it is necessary to make use of the compute power from supercomputers.
In this project that was carried out in the period of May 2012 to April 2013, the software SphäroSim was parallelized with the Message Passing Interface (MPI) and an additional multithreading layer in order to leverage the compute power of a compute cluster for the first time. As Intel’s Xeon Phi architecture was promising with respect to further performance gain, the parallel application was further ported to this Many Integrated Core (MIC) processor. Overall, this project enables the improvement in scaling from original few threads of a desktop computer to over 1000 cores of today’s supercomputers. This allows calculating almost the entire morphology of complex parts.

High-performance clusters often provide multiple MPI libraries and compiler suites for parallel programming. This means that parallel programming tools which often depend on a specific MPI library, and sometimes on a specific compiler, need to be installed multiple times, once for each combination of MPI library and compiler which has to be supported. In addition, over time, newer versions of the tools also get released and installed. One way to manage many different versions of software packages, used by many computing centers all over the world, is the "module" software. However, each center provides a different set of tools, has a different policy on how and where to install different software packages, and how to name the different versions. UNITE tries to improve this situation for debugging and performance tools.