TU Darmstadt / ULB / TUbiblio

An Approach to Visualize Remote Socket Trafic on the Itel Nehalem-EX

Iwainsky, Christian and Reichstein, T. and Dahnken, C. and an Mey, D. and Terboven, C. and Semin, A. and Bischof, C.
Guarracino, M. and Vivien, F. and Träff, J. and Cannatoro, M. and Danelutto, M. and Hast, A. and Perla, F. and Knüpfer, A. and Di Martino, B. and Alexander, M. (eds.) (2011):
An Approach to Visualize Remote Socket Trafic on the Itel Nehalem-EX.
Berlin / Heidelberg, Springer, In: Euro-Par 2010 Parallel Processing Workshops, In: Lecture Notes in Computer Science, [Conference or Workshop Item]

Abstract

The integration of the memory controller on the processor die enables ever larger core counts in commodity hardware shared memory systems with Non-Uniform Memory Architecture properties. Shared memory parallelization with OpenMP is an elegant and widely used approach to leverage the power of such systems. The binding of the OpenMP threads to compute cores and the corresponding memory association are becoming even more critical in order to obtain optimal performance. In this work we provide a method to measure the amount of remote socket memory accesses a thread generates. We use available performance monitoring CPU counters in combination with thread binding on a quad socket Nehalem EX system. For visualization of the collected data we use Vampir.

Item Type: Conference or Workshop Item
Erschienen: 2011
Editors: Guarracino, M. and Vivien, F. and Träff, J. and Cannatoro, M. and Danelutto, M. and Hast, A. and Perla, F. and Knüpfer, A. and Di Martino, B. and Alexander, M.
Creators: Iwainsky, Christian and Reichstein, T. and Dahnken, C. and an Mey, D. and Terboven, C. and Semin, A. and Bischof, C.
Title: An Approach to Visualize Remote Socket Trafic on the Itel Nehalem-EX
Language: English
Abstract:

The integration of the memory controller on the processor die enables ever larger core counts in commodity hardware shared memory systems with Non-Uniform Memory Architecture properties. Shared memory parallelization with OpenMP is an elegant and widely used approach to leverage the power of such systems. The binding of the OpenMP threads to compute cores and the corresponding memory association are becoming even more critical in order to obtain optimal performance. In this work we provide a method to measure the amount of remote socket memory accesses a thread generates. We use available performance monitoring CPU counters in combination with thread binding on a quad socket Nehalem EX system. For visualization of the collected data we use Vampir.

Series Name: Lecture Notes in Computer Science
Volume: 6586
Place of Publication: Berlin / Heidelberg
Publisher: Springer
Divisions: 20 Department of Computer Science
20 Department of Computer Science > Scientific Computing
Zentrale Einrichtungen
Event Title: Euro-Par 2010 Parallel Processing Workshops
Date Deposited: 10 Jul 2012 14:15
Export:

Optionen (nur für Redakteure)

View Item View Item