Parallel application performance issues change with scale and often necessitate large-scale execution measurements for their investigation. Performance measurement and analysis becomes increasingly challenging as the numbers of processes and threads get very large, as the sizes of both profiles and event traces tend to grow proportionally. It is therefore recommended to start from modest scale and then make execution configurations successively larger. Examples will be presented of using the Scalasca toolset that was specifically developed for scalable performance analysis of large-scale parallel applications.
The 30-minute live webinar presented how to understand the challenges of large-scale application executions and the associated performance measurement and analysis issues, discussed best practice use of the Scalasca toolset and how it is used with some large-scale application executions.
The presentation slides are also available here.
Brian Wylie has been a scientific researcher in Juelich Supercomputing Centre of Forschungszentrum Juelich since 2004, in the group developing the Scalasca toolset for scalable performance analysis of large-scale parallel applications. He contributes to the EU Centre of Excellence for Performance Optimisation and Productivity (POP), and management of the High-Q Club of highest scaling codes on the JUQUEEN BlueGene/Q and associated workshops. He also coordinates training activities of the Virtual Institute -- High Productivity Supercomputing (VI-HPS), organising bespoke hands-on training workshops at leading HPC centres as well as tutorials at international conferences.