×

Visualizing the memory access behavior of shared memory applications on NUMA architectures. (English) Zbl 0983.68756

Alexandrov, Vassil N. (ed.) et al., Computational Science - ICCS 2001. International conference, San Francisco, CA, USA, May 28-30, 2001. Proceedings. Part 2. Berlin: Springer. Lect. Notes Comput. Sci. 2074, 861-870 (2001).
Summary: Data locality is one of the most important issues affecting the performance of shared memory applications on NUMA architectures. A possibility to improve data locality is the specification of a correct data layout within the source code. This kind of optimization, however, requires in depth knowledge about the run-time memory access behavior of programs. In order to acquire this knowledge without causing a probe overhead, as it would be caused by software instrumentation approaches, it is necessary to adopt a hardware performance monitor that can provide detailed information about memory transactions. As the monitored information is usually very low-level and not user-readable, a visualization tool is necessary as well. This paper presents such a visualization tool displaying the monitored data in a user understandable way thereby showing the memory access behavior of shared memory applications. In addition, it projects the physical addresses in the memory transactions back to the data structures within the source code. This increases a programmer’s ability to effectively understand, develop, and optimize programs.
For the entire collection see [Zbl 0968.68554].

MSC:

68U99 Computing methodologies and applications
68T45 Machine vision and scene understanding

Software:

SMiLE
PDFBibTeX XMLCite
Full Text: Link