Techniques for adapting industrial simulation software for power devices and networks to multi- and many-core architectures

Meer over het boek

Simulation software has been extensively utilized in both academic and industrial settings for years. Recently, hardware characteristics have evolved rapidly, with multi-core systems becoming commonplace, even in mass-market laptops. Unlike the past, when performance improvements relied on higher processor clock frequencies, existing software often fails to automatically leverage these additional cores and enhancements. To fully utilize today’s multi- and many-core architectures, software must be adapted accordingly. This thesis presents cost-effective techniques for enhancing the efficiency of industrial high voltage engineering applications on these architectures, using real-world simulation software from ABB. The proposed methods involve strategic modifications to data structures and algorithms to improve time complexity, such as integrating caches or refining data access methods. Performance enhancements are also achieved by considering processor hardware characteristics, like caches and branch prediction, ultimately leading to the dynamic generation of optimized, problem-specific code. The thesis evaluates the implications of reimplementing applications based on improved theoretical methods concerning performance and cost-effective development. Additionally, it illustrates various application characteristics and runtime parameters that influence parallel efficiency on both general-purpose multi-core processors and a Xeon