Published Articles >> Table of Contents >> Abstract

18th International Parallel and Distributed Processing Symposium (IPDPS'04) - Papers   p. 35b
Running OpenMP Applications Efficiently on an Everything-Shared SDSM
J. J. Costa, Universitat Politècnica de Catalunya - Barcelona
T. Cortes, Universitat Politècnica de Catalunya - Barcelona
X. Martorell, Universitat Politècnica de Catalunya - Barcelona
E. Ayguade, Universitat Politècnica de Catalunya - Barcelona
J. Labarta, Universitat Politècnica de Catalunya - Barcelona

Full Article Text:

DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/IPDPS.2004.1302950
Send link to a friend

Abstract
Traditional software distributed shared memory (SDSM) systems modify the semantics of a real hardware shared memory system by relaxing the coherence semantic and by limiting the memory regions that are actually shared. These semantic modifications are done to improve performance of the applications using it. In this paper, we will show that a SDSM system that behaves like a real shared memory system (without the afore mentioned relaxations) can also be used to execute OpenMP applications and achieve similar speedups as the ones obtained by traditional SDSM systems. This performance can be achieved by encouraging the cooperation between the SDSM and the OpenMP runtime instead of relaxing the semantics of the shared memory. In addition, techniques like boundaries alignment and page presend are demonstrated as very useful to overcome the limitations of the current SDSM systems.
Back to Top Additional Information
Citation:  J. J. Costa, T. Cortes, X. Martorell, E. Ayguade, J. Labarta, "Running OpenMP Applications Efficiently on an Everything-Shared SDSM," ipdps, p. 35b,  18th International Parallel and Distributed Processing Symposium (IPDPS'04) - Papers,  2004