Extra-P – A performance-modeling tool for applications on HPC systems
It is not uncommon for HPC applications to exhibit poor performance when scaled to larger systems. To address this issue, researchers from TU Darmstadt and ETH Zurich have developed Extra-P, a software tool that enables application developers to identify the sources of poor scaling before it becomes apparent. By automatically learning the scaling functions of individual code regions from performance metrics, Extra-P can predict performance for hypothetical execution configurations, including those with larger processor counts. Extra-P is freely available and easy to use.
The Laboratory for Parallel Programming devises novel methods, tools, and algorithms to exploit massive parallelism on modern hardware architectures. Currently, conducting research in the areas: discovery of parallelism in sequential programs, performance modelling of parallel programs, scalable parallel algorithms, resource management in HPC and cloud environments, deep neural networks.