A Methodology for generating Dynamic Tuning strategies in Multicore Systems

One of the main consequences of current HPC systems heterogeneity is that different levels of parallelism should be considered in all phases of parallel application development. Therefore, support tools and, in particular, performance analysis and tuning tools, must also be adapted to manage heterog...

Full description

Saved in:
Bibliographic Details
Published inProceedings of the International Conference on Parallel and Distributed Processing Techniques and Applications (PDPTA) p. 1
Main Authors Allande, César, Jorba, Josep, Sikora, Anna, César, Eduardo
Format Conference Proceeding
LanguageEnglish
Published Athens The Steering Committee of The World Congress in Computer Science, Computer Engineering and Applied Computing (WorldComp) 01.01.2012
Online AccessGet full text

Cover

More Information
Summary:One of the main consequences of current HPC systems heterogeneity is that different levels of parallelism should be considered in all phases of parallel application development. Therefore, support tools and, in particular, performance analysis and tuning tools, must also be adapted to manage heterogeneity. A significant step forward in this adaptation consists of developing specific strategies, to automatically improve the performance of the parallel regions of the application being executed in each computing element. The authors propose a methodology, to systematically develop performance optimization strategies for specific application patterns taking into consideration hardware characteristics. These performance optimizations are intended to be applied by means of the management code provided by most high level libraries. This study describes the methodology developed and shows how it can be used to expose performance factors that can be dynamically tuned on an OpenMP application.