Implicitly-Parallel Functional Dataflow for Productive Hybrid Programming on Blue Waters

Michael Wilde, University of Chicago

Usage Details

Michael Wilde, Mihael Hategan-Marandiuc, Justin Wozniak, David Kelly, Ketan Maheshwari, Yadu Babuji, Scott Krieder, Timothy Armstrong

As computing systems become increasingly parallel employing hybrid architectures, the difficulty of programming them increases dramatically. The long-sought-after goal in programming model research is to make parallel programming automatic, freeing the programmer from explicitly dealing with parallelism. This project explores a programming model which addresses the urgent problem of how to simplify the programming of complex hybrid systems, through the Swift and GeMTC projects. We believe Blue Waters offers an ideal testbed to explore a data- flow parallel programming system on a petascale hybrid architecture and a diverse set of scientific applications (e.g., glass material modeling, protein structure, and molecular dynamics).

http://www.mcs.anl.gov/~wilde/