ISBN 978-3-8439-5811-0

978-3-8439-5811-0, Reihe Informatik

Alexander Van Craen
Performance Engineering Between Theory and Practice: Algorithmic Design, Hardware-Oriented Optimization, and Their Integration into Higher Education

258 Seiten, Dissertation Universität Stuttgart (2026), Hardcover, A5

Zusammenfassung / Abstract

Performance engineering has become a central challenge in modern high-performance computing, driven by architectural heterogeneity, complex software stacks, and growing demands for efficiency, scalability, and energy awareness. Achieving high performance requires coherent decisions across the computational stack, from numerical methods and data layouts to compiler behavior, hardware characteristics, and measurement methodology.

This dissertation investigates performance engineering through complementary case studies: the ice-sheet model PISM, the distributed sparse-grid framework DisCoTec, the heterogeneous machine-learning library PLSSVM, the hardware sampling library HWS, and N-body simulations used as research benchmarks and teaching instruments. A unified workflow combines baseline measurements, Roofline-based classification, scaling and communication analysis, architecture-aware tuning, and energy-related hardware metrics.

The results show that performance engineering is inherently multi-layered and must be treated as a continuous process. Numerical structure, data movement, communication patterns, compiler behavior, and hardware characteristics jointly determine efficiency and scalability. Runtime and energy efficiency do not necessarily correlate, and algorithmic and architectural choices exhibit characteristic performance and energy signatures.

Beyond individual applications, this dissertation contributes reusable tools, including a vendor-agnostic hardware sampling library and a heterogeneous PLSSVM-based benchmark, and demonstrates how performance-engineering concepts transfer into higher education through N-body-based teaching formats.

Overall, this work positions performance engineering as a holistic discipline integrating numerical methods, software architecture, hardware awareness, measurement methodology, and pedagogy, supporting robust, efficient, and sustainably maintainable scientific software.