Enabling Performance Engineering in Hesse and Rhineland-Palatinate - A Cooperation of the Rhine-Main-Universities and the Technische Universität Kaiserslautern

点击翻译按钮获取中文摘要

The objective of this proposal is to deepen the service scope and to mesh existing structures for basic HPC programming and tuning support at the Technische Universität Darmstadt, the Goethe-Universität Frankfurt, the Johannes-Gutenberg-Universität Mainz (the Rhein-Main Universities), and the Technische Universität Kaiserslautern. Among the four sites represented here, we have both a rich pool of challenging HPC applications emanating from first-class science, but also researchers well-established in critical performance-relevant research that we want to capitalize on.Thus, we want to provide more depth in HPC service in areas where scientific expertise at these universities coincides with critical user needs, pushing expertise on GPU performance engineering, model-based scalability analysis, and algorithmic stability and reproducibility into performance engineering practices. The issue of GPU and vectorization performance is crucial, as it is relevant for all HPC architectures. Performance modeling is critical, as without it, we cannot identify potential performance bottlenecks before resources are expended unnecessarily, an issue that is crucial for the scaling-up of applications, and, in particular, the efficient transition between tier-3, tier-2, and tier-1 computing centers. Lastly, the issues of algorithmic stability, performance, and reproducibility are becoming more and more important with respect to the plausibility and credibility of HPC simulations. So this choice of topics is by no means random but deliberate; the scientific expertise of the project team dovetails with real HPC user needs.At a governance level, we want to bundle the distributed expertise for HPC support and performance engineering that has been already established within the Hessian HPC-Competence Center and the Alliance for HPC in Rhineland-Palatinate in a new umbrella organization for the HPC support and performance engineering activities of all universities of Hesse and Rhineland-Palatinate. In particular, we will ensure dedicated on-site basic level support at all sites, and mesh the postdoctoral associates that provide the translational effort described above with already-existing HPC programming and tuning staff, thus creating a vibrant distributed group of scientific staff that has a shared fundamental knowledge of programming and tuning as well as the topical capabilities mentioned above. We will provide tutorials, coding workshops, and capability showcases to push our special capabilities. To show the value of this work, we will define success metrics, and document and evaluate the work performed.We believe that in this fashion, we create a service organization that offers organizational and operational synergies for HPC users in our states, and profits from the close spatial proximity of the sites involved. In addition, we will provide a model for meshing basic on-site HPC support in a distributed organization with focus support for topics of relevance.