Performance analysis in the Petascale era

Simulations on HPC systems have become an indispensable key technology in modern science and engineering. However, to cope with current trends in HPC hardware technology, applications are required to efficiently exploit the available parallelism using either shared- or distributed- memory parallelization strategies, or even combinations of both. This talk will motivate the need to apply performance analysis tools during the development of simulation codes to achieve good performance and scalability. After providing a brief overview of commonly used analysis techniques, two state-of-the-art open-source toolsets will be introduced in more detail: the community instrumentation and measurement system Score-P as well as the performance analysis toolset Scalasca. Finally, case studies will demonstrate the usefullness of these tools and outline current research activities