SC²S Colloquium - November 25, 2016
|Date:||November 25, 2016|
|Time:||3:00 pm, s.t.|
Fabian Güra: A Generic Profiling Infrastructure for the Hyperbolic PDE Solver Engine ExaHyPE
The ExaHyPE project is tasked with tackling the software side challenges of exascale high performance computing. In order to optimize and evaluate the novel algorithmic and numeric approaches developed as part of the project, metrics correlated with performance and energy efficiency need to be obtained. In this thesis we therefore discuss design goals, architecture and exemplary implementations of a generic profiling infrastructure for the hyperbolic PDE solver engine. Against the background of hyperbolic balance laws and a state of the art limited ADER-DG scheme we present preliminary profiling results to illustrate the employed techniques. In two case studies we demonstrate how the generic profiling infrastructure can be used to quickly track down the source of possible issues in a simulation without changing a single line of code and how it provides guidance for optimization efforts within the scope of metrics-driven performance engineering.