6238

Run, Stencil, Run! – A Comparison of Modern Parallel Programming Paradigms

Helmar Burkhart, Matthias Christen, Max Rietmann, Madan Sathe, Olaf Schenk
University of Basel, Department of Mathematics and Computer Science, Klingelbergstrasse 50, CH-4056 Basel, Switzerland
Workshop on Parallel Systems and Algorithms (PARS), 2011

@article{burkhart2011run,

   title={Run, Stencil, Run!},

   author={Burkhart, H. and Christen, M. and Rietmann, M. and Sathe, M. and Schenk, O.},

   year={2011}

}

Download Download (PDF)   View View   Source Source   

703

views

While the performance of supercomputers has increased dramatically during the last 15 years, programming models and programming languages have more or less remained constant. Two de facto standards, the Message Passing Interface (MPI) for programming distributed memory architectures and OpenMP for programming shared-memory architectures still dominate the field of computational science and engineering. As current supercomputers are constellations with symmetric multiprocessor nodes that are coupled by high-speed inter-connects such as Infiniband, hybrid versions OpenMP/MPI have been introduced. From the programmer’s point of view, this is rather complicated and the resulting software productivity is not at all comparable with corresponding practice of commercial software elsewhere. While today’s desktop computers are already parallel systems, this effect will be even more accentuated in the coming years. Extrapolation from the current TOP500 list tells us that a desktop or laptop computer will have more than 1000 processor elements in 10 years from now. Thus, in order to use computers efficiently, all programmers today have to address parallelism issues.
No votes yet.
Please wait...

* * *

* * *

HGPU group © 2010-2017 hgpu.org

All rights belong to the respective authors

Contact us: