A Unified HPC Performance Measurement Infrastructure – Overview and Results of the BMBF Project SILC

Andreas Knüpfer¹, Kai Diethelm²

¹ Technische Universität Dresden
² GNS mbH Braunschweig

 (slides)

In todays HPC environments with heterogeneous hardware and complex software
the processes of performance optimization and scalability enhancements become
increasingly difficult. At the same time, they become more and more
indispensable to reach satisfyingly efficient resource usage.

Therefore, adequate performance analysis tools are an essential part of the
HPC software stack. The tools themselves face the same challenges as HPC
applications, in particular in terms of scalability to keep pace with the
hardware and software evolution.

For the sake of tool users as well as the tool's own technological progress, a
number of established groups in that field joint their efforts in providing a
high-quality performance measurement infrastructure under the roof of the BMBF
project SILC (Skalierbare Infrastruktur zur automatischen Leistungsanalyse
paralleler Codes).

The talk will present the initial goals of project as well as the almost final
results. This will include the mutual benefits of the joint measurement
infrastructure, the design and the main components, evaluation results, and an
overview about a typical usage scenario from industry. Furthermore, there will
be an outlook on the continued collaboration beyond the project's funding
period and spin-off activities.