Computer Science

Programming Challenges Workshop

Programming Challenges
July 27 - 29 2011
Marina del Rey, CA
 
   ADDITIONAL INFORMATION
Workshop Challenges »
Agenda .pdf file (87KB)
Introduction - Day 1 .pdf file (124KB)
Introduction - Day 2 .pdf file (115KB)
Abstracts & Biographies »
Presentations »
List of Attendees, Speakers and Panelists .pdf file (45KB)
Workshop Report .pdf file (6.0MB)
WORKSHOP COMMITTEE:
Saman Amarasinghe (MIT) Mary Hall (U. Utah)
Pat McCormick (LANL) Richard Murphy (Sandia)
Keshav Pingali (U. Texas-Austin) Dan Quinlan (LLNL)
Vivek Sarkar (Rice) John Shalf (LBNL)
 
ORGANIZING COMMITTEE:
Bob Lucas Kathy Yelick
 
ASCR POINTS OF CONTACTS:
Sonia R. Sachs (bio).pdf file (77KB) Lenore Mullin (bio).pdf file (43KB)
 
WORKSHOP GOALS:

This workshop will explore the demands on programming models, languages, compilers and runtime systems for Exascale machines and identify promising approaches to meeting them and measuring their progress. A goal will be to define objective criteria for assessing programming models and language features that enable effective use of diverse exascale architectures for important science applications.

The workshop will consider both evolutionary and revolutionary solutions for Exascale programming challenges. For the evolutionary approaches, MPI can be used as the software substrate for inter-node communication and the programming challenges will need to focus on novel hybrid approaches that integrate intra-node parallelism with MPI communications. This integration may result in programming models that do not expose MPI directly to the programmer, but use MPI under the covers for inter-node communication akin to other system services like memory management. For the revolutionary approaches, the software substrate for inter-node communication can be different from MPI for superior integration with multi-threaded nodes. There can be a natural migration path for applications from the evolutionary to revolutionary approaches, as inter-node communication becomes more tightly integrated and unified with intra-node parallelism.

At the end of this workshop we will have clear articulation and prioritization of challenges (what we know, and what remains uncertain), prioritized list of responses, roadmap with timelines for implementing those responses, and a rough idea about the order of magnitude of costs involved.

A report that summarizes the workshop discussions and provides recommendations on the programming directions will be generated by the workshop committee. The final report will be delivered by September 15, 2011.

Last modified: 3/15/2013 5:35:19 PM