SES-HPC: Services for Experienced and Starting HPC Tier 3 users

• Applicant: Professorin Dr. Sabine Roller, until 2/2021
• Subject Area: Data Management, Data-Intensive Systems, Computer Science Methods in Business Informatics ; Software Engineering and Programming Languages
• Term: from 2016 to 2022
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 320892061

A transparent usage of the entire HPC pyramid requires support and training of users on all levels. Unfortunately, we observe a strong discrepancy in the large number of available offers on the higher level (Tier-1, Tier-0), and less support the further down on the pyramid we go. But HPC competences need to be gained by the users as well as by their software on Tier-3 level where we can first catch them up on the local university level by personal interaction. Thus, this project proposes an incentive-driven approach for services aiding HPC users on Tier-3 level. We concentrate on experienced users who write their own codes and are on the verge to Tier-2 and Tier-1 systems. On the one hand, performance review will be offered. In this review the performance, efficiency and scalability of the code will be analyzed together with the scientists. Identified problems (e.g. poor serial performance, weak scalability, or inappropriate numerical algorithms etc.) shall be tackled by suggesting improvements. Main focus will be laid to those areas of the code where the highest impact of the code change onto the performance characteristics of the code can be achieved. To ensure that the achieved improvements can be used and maintained by the users even in the future, the second pillar of the project is teaching and training. Courses that teach basics of performance and parallelization the scientists shall prepare for moving their software to the next higher Tier if necessary. As a third pillar, a small amount of effort will be spent to support those users with intermediate experience, typically employing commercial or open-source software to solve their scientific problems at hand. We offer support both in finding optimal settings for the given software on the current hardware (e.g. optimal number of MPI processes to number of OpenMP threads) and finding the most suitable hardware for the problem and help the user getting access to it. E.g. OpenFoam for fluid dynamics simulations is available on Tier-1 systems at HLRS, chemical software like NWchem is available at Tier-2 system in Cologne. This step up the pyramide is the forth pillar of the project: supporting the Tier-3 users in applying for higher levels. This includes finding the best suited system, testing and evaluating the software on this system as well as support in administrative issues like where to find the proposals and how to fulfill the requirements. An important aspect in the project will be an organized knowledge transfer to ensure a sustainable environment of services for HPC users. Result workshops will be held where a broader audience is targeted (e.g. scientists from other centers using the same software, industry partners interested in how their programs are used, etc). Additionally the results of performance or parameter optimizations will be documented in a wikipedia-style portal which will be available to everybody.

• DFG Programme: Research Grants
• Co-Investigator: Daniel Harlacher