The documents distributed by this server have been provided by the contributing authors as a means to ensure timely dissemination of scholarly and technical work on a noncommercial basis. Copyright and all rights therein are maintained by the authors or by other copyright holders, notwithstanding that they have offered their works here electronically. It is understood that all persons copying this information will adhere to the terms and constraints invoked by each author's copyright. These works may not be reposted without the explicit permission of the copyright holder.
Publications of SPCL
|An Overview of Static & Dynamic Techniques for Automatic Performance Modeling|
(Presentation - presented in Frankfurt, Germany, Jun. 2016, Invited talk at International Supercomputing Conference )
AbstractAutomatic performance modeling of high-performance computing applications is a quickly emerging research field. The goal is to automatically derive analytical equations that model the resource consumption or execution time of arbitrary applications. These equations can then be used to determine performance bottlenecks which may be classified as bugs by the user, to derive optimal execution parameters such as numbers of processes or mappings, or to co-design computing systems, among many others. Techniques for automatic performance modeling broadly fall into two categories: static and dynamic methods. Static methods involve source-code analysis and can deliver provable performance bounds, yet, they cannot handle arbitrary codes. Dynamic techniques often build on statistical methods to derive performance models from observations which may which may not exactly capture the true performance model. In this talk, we describe and exemplify both methods and show how to apply them in practice.