Computer Science Colloquium
University of Passau
Analysing the Performance Evolution of Run-Time Configurable SystemsWed 23.01.2019, 10:00, 60 minutes
Informatikgebäude, S3 218
AbstractWith the high number of configuration options available, finding an optimal configuration of a highly configurable system for a certain task is nearly impossible and even finding a good one is difficult. Often the impact a configuration option has on performance is not clear and even less known are the different performance interactions between configuration options. Furthermore, software systems change over time. An optimal performing configuration today could become an underperforming one tomorrow.
To improve our understanding of how highly configurable software systems evolve and to tackle this particular problem, we have been developing Variability-aware Region Analyzer (VaRA). VaRA uses control-flow and data-flow analyses to find regions in code that relate to run-time configuration options. It is built on the LLVM compiler infrastructure and can be used automatically during the compilation of a program. The identified configuration-dependent regions then get instrumented with measurement code to generate performance data that are linked to configuration options (i.e., their influence on performance). By executing a specific benchmark or test suite with a binary modified by VaRA, we can build a fine-grained performance model. Generating these models continuously for all revisions allows us to monitor performance changes over time.
BioFlorian Sattler is a PhD student at the Software Engineering Department of the University of Passau (Prof. Sven Appel). His research interests are in Performance Analysis/Prediction, Evolution of Software, and Software Product Line Engineering.
Invited by Dr. Manuel Rigger, Institut für Systemsoftware
The Computer Science Colloquium is organized by the Department of Coputer Science at JKU, the Österreichische Gesellschaft für Informatik (ÖGI) and the Österreichische Computergesellschaft (OCG).