Aftermath: Performance analysis of task-parallel applications on many-core NUMA systems
Andi Drebes, Karine Heydemann, Nathalie Drach, Antoniu Pop, Albert Cohen · HAL (Le Centre pour la Communication Scientifique Directe) · 2014
We show how to analyze the locality of memory accesses using Aftermath, an open source tool for interactive, off-line trace visualization and performance analysis of task-parallel programs and run-time systems executing on multi-core and many-core systems. In order to help users detect performance anomalies and pinpoint the exact cause, Aftermath offers adequate visual representations of dynamic events for informal exploration as well statistical tools and filters that provide accurate values for performance indicators. In this paper we give an overview of the capabilities of Aftermath to analyze the locality of memory accesses with respect to NUMA.