1. 6thWorkshop of the INRIA-Illinois Joint Laboratory for PetaScale Computing November 21-23 2011 NCSA, Urbana, IL

2. –Between INRIA and UIUC –In the context of Blue Waters –From 2009 to end of 2013 –2 co-executive directors: Marc Snir (ANL, UIUC) and Franck Cappello (INRIA) –3 types of activities: workshops, visits, missions + Collaborations: FR, EU, America, ASIA –Productions: Publications and Software –Google “joint-lab”  first link INRIA-Illinois Joint-Laboratory

3. INRIA-Illinois Eco system JLPC Users (Center for Atmospheric research, Japan Earth quake, etc.) Application developers System software R&D (Charm++, HDF5, Integrated System Consol, FT-MPI, etc.) Algorithm research and development (Numerical methods, Mapping + load balancing, FT, I/O) Hardware vendor (CRAY) System practitioners. HPC System Applications (PRAC): -Contagion on very large social networks -Earthquake system science -Formation of the first galaxies ETC.

4. Joint-Lab Areas of Collaboration Numerical Algorithms and Libraries (Communication avoiding, dynamic scheduling, resilient methods for graph partitioning, linear algebra, FFTs, etc.); Programming and runtime (parallel languages, runtime, multi-threading, process mapping, scheduling, communication libraries, etc.) Resilience (FT-protocols, event log analysis, diskless checkpointing, resilience performance modeling, energy consumption, etc. ); System software (monitoring, resource-job manager, I/O environement, archive, etc.).

5. INRIA-Illinois Joint-Laboratory 3D-FFTTornado SimulationNew fault Tolerance Approaches « Application driven »: From Applications to system software Applications Numerical librariesSystem Software New I/O Approaches Earth Quake Simulation Climate Simulation New Archiving Approaches Linear Agebra

6. Joint-Lab International collaborations Japan Canada NCSA Brasil Germany Spain INRIA UTK

7. 6th Workshop Highlights Session 1 Sustained Petascale The new Blue Water technology Session 2 From Petascale to Exascale Activities at Argonne National Laboratory Session 3 System Software I/O optimization, Archiving policies, Resource/Job manager, Performance modeling Session 4 Numerical Algorithms and Libraries Multi-Grid Preconditioners, Communication avoiding LU and QR, Graph partitioning, 3D-FFT Session 5 Programming and runtime environments Programming and runtime for GPUs, Optimizations for Charm++, Dataflow overhead

8. Breakout session Wednesday Morning: 3 hours of open discussions Small groups and focus on specific topic More like a brainstorming meeting. Try to resolve some research issues or to plan for it. Define how the collaboration will be implemented in the next 6 months. The research results will be presented at the next workshop Try to implement collaboration with visits. Visits could be in two directions: to UIUC and to France

9. Workshop Logistics 1/3 Monday and Tuesday sessions on Auditorium Wednesday Break out sessions start on Auditorium We split on Friday morning for breakout and reconvene at 12:00 before lunch for report and closing For any question, please ask Beth or Judy at the lobby

10. Workshop Logistics 2/3 All lunches will take place in the NCSA atrium Breaks will take place in the NCSA atrium Diners will take place in different restaurants Two mini buses (driven by Marc and Franck) will pick you up at the Hampton 1/2 hour before the diner (as mentioned in the program)

11. Workshop Logistics 3/3 Use your laptop We collect ALL presentations Presentations will be stored on the join-lab web site. Please put you presentation on the USB key, right after your talk.

12. Acknowledgments Invited speakers who accepted to attend the Workshop and present their latest work Rajeev Thakur, ANL Robert Ross, ANL Paul Hovland, ANL George Bosilca, UTK Daisuke Takahashi, University of Tsukuba

13. Acknowledgments

14. Enjoy the Workshop!

15. Break out sessions

16. Break out session help for : -defining objectives for the next 6 months: -joint publications, software, ANR/NSF proposals -identification of how to implement the collaboration -identification of new collaboration topics -addressing immediate technical questions

17. Breakout session Long visits (3 months or more) of students is a very productive way to implement a collaboration  This is the perfect time to propose internships for masters in France  During the break out, try to establish Internship title + a short abstract

18. Breakout session 2 sessions, 6 groups for now. New groups may be organized if needed Objective for each group (each group leader will report on): -Focus on 1 to 3 research project(s): -Define clear and reachable objectives: -conference paper submission -piece of co-developed software -submission of joint proposals -Break down the essential steps toward each objective -Internship proposal -define visits (students and faculties), schedule  next workshop is in 6 months.

19. Breakout session 1 9:00 – 10:15 Room 1030 Routing, topology mapping, scheduling, perf. Modeling Marc Snir Hoefler, Vivien, Gautier, Kale, Namyst, Méhaut, Bohm, Pilla, Amedo, Perez, Baboulin Room 1040 Resilience Kramer Cappello, Gainaru, Ropars, Menese, Bautista, Antoniu, Richard, Fourestier, Jacquelin

20. Breakout session 2 10:30-12:00 Room 1030 Programming Models GPUS Kale, Méhaut, Namyst, Wu, Amedo, Perez, Bohm, Pilla, Baboulin, Fourestier, Gautier Room 1040 I/O Snir, Viven, Jaquelin, Antoniu, Richard, Kramer, Gainaru, Ropars, Cedric Mathieu Room 1104 3D-FFT Cappello, Takahashi, Yee, Jeongnim, Hoefler

21. Breakout session  Sponsoring a student for a visit in France:  Bourses Chateaubriand de l'Ambassade de France, pour l'accueil en France de doctorants en cours de thèse dans une université américaine (avec priorité aux thèses en cotutelle, ou pouvant le devenir) et en Sciences exactes et appliquées, Sciences de la vie et de la santé (date limite 31 janvier 2011) http://france-science.org/chateaubriandhttp://france-science.org/chateaubriand  Bourses de la Commission franco-américaine, pour différents échanges: http://www.fulbright-france.org, en particulier pour l'accueil aux Etats-Unis, sur financement MESR-Département d'Etat, de doctorants en cours de thèse dans une université française : http://www.fulbright- france.org/gene/main.php?base=343 (date limite 1er février 2011). http://www.fulbright-france.orghttp://www.fulbright- france.org/gene/main.php?base=343  Partner University Fund (PUF), Ces collaborations prennent diverses formes le plus souvent complémentaires : recherche et publications, formations communes, double diplômes aux niveaux master et doctorat. Elles peuvent inclure des opportunités de mobilité pour les étudiants, post-doctorants, professeurs et chercheurs, et peuvent dotées à hauteur de 80 000 USD par an, sur trois ans. Les dossiers de candidatures et les critères de sélection se trouvent sur le site de PUF : http://www.facecouncil.org/puf, avec une date limite fixée au 15 décembre 2010.http://www.facecouncil.org/puf

22. Closing the Workshop

23. Before Closing Many thanks to NCSA people who helped in the preparation and organization of the meeting Tech Support George Estes Michael Miller Paul Ponder Public Affairs Bill Bell Director's Office Beth McKown Judy Olson

24. Closing the fourth Workshop Very successful –More than 50 attendees at peak –Presentation of results and early result on existing collaborations –Break-out session worked pretty well Thanks for your very active participation

25. Acknowledgments

26. The 7 th Workshop of the INRIA-Illinois Joint Laboratory On PetaScale Computing June-July 2012, Rennes, Britany

27. 3D-FFT 2 different problems: –BW benchmark: Multiple 3D-FFT (UPC may be), 12K size, etc. –Library (force the memory layout or get the memory layout)  Definition of the interface:Kim+Daisuke+Robert) –Multi-algorithms (decompisition) in the library (selection based on modeling)  Daisuke –Performance modeling: Alex, Daisuke + Kim + Torsten + Hormozd –Strategies for overlapping communication and computation  Daisuke + Camille + Torsten –Implementation of the real to complex  Daisuke. –Performance comparison between P3DFFT an FFTE using Bigsim + Gengbin Zheng; Ehsan Totoni In 6 months : 10% efficiency on 1536^3 on 512 cores 10% effciency on 2K^3 on 1K cores Implementation of the collaboration  weekly meeting: Alex, Kim, Torsten, Camille, Daisuke, Franck + others.