The Current Challenges
in DataFlow Supercomputer Programming

ISCA

Tel Aviv, June 24, 2013










Team

Michael J. Flynn

Member of the Board, Maxeler

Professor, Stanford University

Oskar Mencer

CEO, Maxeler

Professor, Imperial College London

Itay Greenspon

Maxeler,

London, England, UK

Veljko Milutinović

University of Belgrade

Member of the SAB, Maxeler

Šaša Stojanović

University of Belgrade

Ph.D. Student

Živojin Šuštran

University of Belgrade

Ph.D. Student

Goran Rakočević

Mathematical Institute, Belgrade

Ph.D. Student

Dragan Bojić

University of Belgrade

Professor

Milo Tomašević

University of Belgrade

Professor

Miroslav Bojović

University of Belgrade

Professor

Overview

First, the modern programming paradigms are discussed. Second, a novel classification of existing approaches to supercomputing is presented. Third, each existing approach is analyzed using the same template:
(a) A figure explaining the architecture used and
(b) Discussion of major characteristics, pros and cons.
Forth, after a brief overview of MultiCore, ManyCore, CoareGrain DataFlow and FineGrainDataFlow programming models, most of the slieds are dedicated to programming of dataflow supercomputers in general and the Maxeler dataflow supercomputer in particular. Fifth, the essence of the recent paper (by a subset of coauthors of this tutorial) in Communications of the ACM is presented. Sixth, architectural details and programming examples are given for applications in GeoPhysics and Banking.

Background

Flynn, M., Mencer, O., Milutinovic, V., et al
"Moving from petaflops (on simple benchmarks) to petadata per unit of time and power (on sophisticated benchmarks)"
Communications of the ACM, March 2013.

Stojanovic, S. et al,
"A Survey of Modern Reconfigurable Computing,"
Proceedings of the IEEE ICIT Conference,
Athens, Greece, March 19-21, 2012.

Mencer, O., Flynn, M,
"The Maxeler Exascale Computing,"
Maxeler,
London, UK, April 2012.

Slides

Slides for tutorial available here.