Work

Techniques for adapting Industrial Simulation Software for Power Devices and Networks to Multi- and Many-Core Architectures

Year: 2014
Type: dissertation
Abstract: Simulation software has been widely used in academic and industrial environments for a long time.In recent years, however, the available hardware characteristics have changed significantly and rapidly.Several years ago, true parallel processing was only available using clusters or expensive workstations, whereas today systems with multiple processor cores are well established and even mass market laptops provide multiple cores.Contrary to performance increase due to rising processor clock frequency in the past, existing software typically does not automatically benefit from these additional cores and several other improvements.Instead, it has to be adapted to properly benefit from the increased compute power and to efficiently use today's multi-and many-core architectures.This thesis devises techniques to cost-effectively conduct these adaptations to increase the efficiency of industrial high voltage engineering applications on multi-and manycore architectures and help to leverage their full potential.This is done using several real world exemplary simulation software packages provided by industry partner ABB.The presented techniques include strategic changes to data structures and algorithms to improve time complexity by subtly inserting caches or extending data access methods where appropriate.Other changes improve performance by accounting for characteristic properties of processor hardware, such as caches or branch predictionultimately by dynamically generating optimized and highly problem specific code.Additionally, the implications of reimplementing an application based on new and improved theoretical methods regarding performance as well as cost-effective development are discussed and evaluated.Finally, various application characteristics and runtime parameters affecting parallel efficiency are illustrated on both general purpose multi-core processors as well as a Xeon Phi system representing a many-core architecture.Many people have contributed to the success of this work.First of all, I would like to thank Prof. Dr. Arndt Bode for giving me the opportunity to do my doctorate at his chair, the Lehrstuhl für Rechnertechnik und Rechnerorganisation (LRR), for his continuous support and the freedom he granted me for my research.I am also grateful to Prof. Dr. Carsten Trinitis, who supervised my work, provided valuable feedback and directions.Furthermore, I am thankful to my colleagues for many fruitful discussions, leading to new ideas and perspectives, as well as an enjoyable work environment.I am also grateful for the cooperation with ABB, which allowed me to conduct my research on real life industry applications and environments.In this context, I specifically want to
Author: Thomas Müller
Language: English
Open Access status: green
OpenAlex ID: W2749897800
Cites: 52
Cited by:
Related to: 10