Pre-workshop panel on the science case for ultrascale computing in the Office of Science

1. Pre-workshop panel on the science case for ultrascale computing in the Office of Science Washington, DC 23-24 April 2003 This slide set is based on the original pacing presentation, modified by discussion that took place in committee.

2. Happy Planck’s Birthday! • Born:23 April 1858, Kiel • Nobel Prize in Physics, 1918 • Fellow, Royal Society, 1926 • President, Kaiser Wilhelm Gesselschaft (the “German NSF”), 1930-37 & 1945-46; Max Planck Institutes in various fields in Germany today • Died: 4 Oct 1947, Göttingen “An important scientific innovation rarely makes its way by gradually winning over and converting its opponents: it rarely happens that Saul becomes Paul. What does happen is that its opponents gradually die out, and that the growing generation is familiarised with the ideas from the beginning”. - Planck

3. Charge (April 2003, Polansky) • “Identify rich and fruitful directions for the computational sciences from the perspective of scientific and engineering applications” • Build a “strong science case for an ultra-scale computing capability for the Office of Science” • “Address major opportunities and challenges facing computational sciences in areas of strategic importance to the Office of Science” • “Report by July 30, 2003” • “Foster additional workshops, meetings and discussions on specific topics that can be identified and analyzed over the course of the next year” • Though our formal charge is Office of Science-centric, the report may ultimately be combined with others in a multiagency thrust, and should be prepared with a broader science case and need not be confined scientifically to Office of Science interests alone

4. Who we are Plus ex officio members from DOE HQ, etc.

5. Choice of strategies • The “zero sum” strategy: • Try to look better than the other guys in the same program • Improve the appearance of your science and/or denigrate the prospects for theirs • Insist that nothing that now exists can be “on the table” • The growth strategy: • Create a united and confident appearance for the program • Find common ground (or, even better, synergism), as an organizing principle for growth in the program • Allow any necessary specialization for each constituent (but explain it well, so that it extends the program rather than distracting from it)

6. Goals for Wednesday • Not to help SC decide directly what to ask for in FY 2005; rather: • To outline a report for the science-based case for an ultrascale simulation capability in the Office of Science • To plan the structure, schedule, core invitee list, and advance homework assignments of a workshop in June that will provide content for the report • To begin to assign responsibilities for report sections and corresponding workshop components • To delegate to a small core group the detailed subsequent coordination of the workshop (weekly telecons)

7. Goals for Thursday • To digest the minutes of Wednesday in a smaller group with DOE HQ staff • To write up a short summary that can be appended to workshop invitations

8. Envisioning the report • Dozens of authors from labs and universities • Editorial oversight by small board, including expertise from each Office • Written hierarchically, with easy-to-read summaries for Congressional staffers and in-depth sections to impress non-computational experts from scientific community • Broad and as comprehensive as possible, as to areas touched • Deep in a number of illustrative places, through case studies • Written with strong awareness of antecedents and distinct mission • Laden with good charts and good (but not gimmicky) scientific eye-candy

9. Possible topics for report (not in appearance or priority order) • Discussion of relevant previous studies, their impact, and the particular contribution of this report, in context • Discussion of related initiatives that are on-going in other agencies, and in other countries, and perhaps discussion of impacts in the commercial and industrial IT worlds, beyond fundamental research • Discussion of historical outcomes of heavy investment in large-scale computing hardware (e.g., NSF NPACI, ASCI, NERSC, Earth Simulator) • Estimation of the “knees” in the curves of scientific results as a function of increasing power and resolution, for a variety of important applications (climate, astrophysics, QCD, combustion, magnetically confined fusion, proteomics, etc.)

10. Possible topics for report (not in appearance or priority order), cont. • Demonstration of scaling through quantitative performance models, based on code structure, architecture, and programming model • Discussion of algorithmic and software bottlenecks in the motivating applications for the next 5 years and 10 years • Discussion of what makes a balanced program to produce apps that scale as needed and enlarge the computational community to exploit and interpret them (e.g., software, networking, algorithmic research, staffing, training) • Discussion of balance in the hardware architecture and evaluation of positive and negative trends in the marketplace (e.g., number of functional units, number of processors per node, memory BW per processor, caches and/or vectors, network topology/latency/BW, I/O, archival storage)

11. Possible topics for report (not in appearance or priority order), cont. • Discussion of meaningful scientific benchmarks for real apps of all important types (e.g., Eulerian PDEs, Lagrangian PDEs, lattice computations/cellular automata, mixed particle/continuum methods, short- and long-range force MD and N-body problems, bio-informatics string-matching algorithms) • Estimation of community needs, distinguishing between development, capability, and capacity purposes • Discussion of actions that we may need to take to encourage vendors to play with us in building scientific instruments rather than just networked webservers

12. Workshop planning • Selected plenaries on context and structure • Selected plenaries on scientific content and scaling analyses • Parallel sessions of breakout groups organized by: • Disciplinary areas (one session, w/approx. 10 breakouts) • Cross-cutting discussions (two sessions) • Parallel sessions scribed by invited and prepared graduate students; overall meeting and written products professionally scribed with DOE HQ support • Web supported (with a priori and a posteriori postings of position papers and presentations)

13. Beyond the report • Book-length treatment? • Dynamic hypertext document?

14. Antecedents • 1982: “Large Scale Computing in Science and Engineering” (convened by NSB) • Four major recommendations: • Increase access to regularly upgraded supercomputing facilities via high bandwidth networks • Increase research in computational mathematics, software, and algorithms • Train people in scientific computing • Invest in research on new supercomputer systems • Led to establishment of NSF supercomputer centers interconnected by high speed network, in 1984

15. Antecedents, cont. • 1992: “Grand Challenges: High Performance Computing and Communications” (OSTP) • Proposed 30% increase in federal support of HPCC (to $638M/yr) • Four major components: • High performance computing systems • Advanced Software Technology and Algorithms • National Research and Education Network • Basic Research and Human Resources • Two famous charts (Fig 2 on Grand Challenge requirements and expected year of achievement, and Fig 4 on the joint contributions of algorithms and hardware) • Eight one-page case studies (weather, aero design, oncogenes, etc.)

16. Antecedents, cont. • 1993: “From Desktop to Teraflop: Exploiting the U.S. Lead in High Performance Computing” (NSF) • Introduced the computing pyramid • Fourteen recommendations for NSF • Four major challenges: • Remove barriers to rapid evolution of HPC technology • Provide scalable access to all levels of HPC capability • Create incentives to promote broadening of base of participation in HPC • Create intellectual and managerial leadership for the future of HPC • Several appendices, including usage statistics and demographics • Statements by individual panel members

17. Antecedents, cont. • 1995: “Report of the Task Force on the Future of the NSF Supercomputer Centers Program” (NSF) • Ten-year review of the NSF supercomputer center program • Led to the creation of the PACIs • Seven recommendations for NSF • Continue strong, viable centers program • Provide national lead-edge sites with balanced resources for rapid advancement of computational science & engineering • Assure partnering with other major research centers • Announce new competition, on five-year cycles with favoritism for existing sites • Provide research support related to center missions, but not for independent research • Increase involvement of NSF directorates in service unit allocations • Provide leadership in interagency plans for apex systems • Extensive usage data and testimonials by prominent scientists

18. Antecedents, cont. • 1999: “Information Technology Research: Investing in OurFuture” (PITAC) • Issued at peak of economic boom and assigned large share of credit for the new economy to past federal investments in IT, citing “spectacular return” • Findings and recommendation • Federal IT R&D investment is inadequate • Federal IT R&D is too heavily focused on near-term problems • Recommendation: Create a strategic initiative in long-term information technology R&D • Put strong focus on software • Emphasized essential role of government in IT, though more than the half the panel was from industry • Recommended $1.3B/yr increase in federal IT R&D by 2004

19. Antecedents, cont. • 2000: “Scientific Discovery through Advanced Computing” (DOE SC) • Successor to “Scientific Simulation Initiative” based on “National Workshop on Advanced Scientific Computing” (NAS, 1998) • Identified major challenges in each of the five Offices that can only be addressed through advanced in scientific computing • Recommended three research foci: • Scientific challenge codes • Computing systems and mathematical software • Collaboratory software infrastructure • Recommended scientific computing hardware infrastructure • Flagship computing facility • Topical computing facilities • Consciously addressed relationship to ASCI and other federal supercomputing programs

20. Antecedents, cont. • 2003: “Revolutionizing Science and Engineering through Cyberinfrastructure” (NSF) • Presented cyberinfrastructure (CI) requirements of a knowledge economy in analogy to infrastructure requirements of an industrial economy • Overarching finding • IT has crossed thresholds that make a comprehensive cyber infrastructure on which to build new types of scientific and engineering knowledge environments and organizations possible • Overarching recommendation for NSF • Establish and lead interagency Advanced CI Program (ACP) to create, deploy, and apply CI in ways that radically empower all scientific and engineering research and allied education, with new NSF investment of $1B/yr and coordinated co-investment from other agencies • PACIs to be replaced, following two-year extension to preserve human assets of these centers • Presented archetypal balanced 60 Tflop/s machine costing $180M in 2004; need “sufficient number”of these (~5)

21. Lessons from antecedents • Workshops such as this can really make a difference, motivating federal investments for years to come • Not all of these antecedent reports had a similar charter as ours; some were official federal advisory committees • There are many choices of presentation modes within these reports: formal text, side bars, charts, case studies, portfolio profiles, detailed appendices, personal statements, etc., that we can consider using, with the aim of featuring strong cases, while being inclusive of science that is not as ripe at the time of writing, but has important potential

22. Web resources at DOE SC • Ultrascale Simulation for Science (www.ulstrasim.info/doe_docs/) • 12 “final release” documents (8/02-4/03) • 16 “working” documents (8/03-4/03) • Scientific Discovery through Advanced Computing • 76 “2-pagers” prepared for the 2003 SciDAC PI meeting (3/03) (http://www.osti.gov/scidac/updates2003.html) • Proceedings (http://www.nersc.gov/conferences/SciDAC2003) • Recent whitepapers precusory to new initiatives throughout the Offices • Theory and Modeling in Nanoscience • Genomes to Life • Fusion Simulation Project • etc. • The “applications” matrix (www.appsmatrix.info)

23. Other resources • Topical computing center whitepapers • Blue planet and Cray X1 proposal and evaluation documents • Strategic plans of the individual Offices • Advisory committee reports

24. Related initiatives in progress • OASCR strategic plan (Dan Hitchcock) • “Mission Computing” workshop (Gary Johnson) • Hardware taskforce (Rick Stevens) • ASCAC HPC roadmap taskforce (Greg McRae) • NITRD “High End Computing Revitalization Task Force” (HECRTF) (www.itrd.gov/hecrtf-outreach/); see also HPCWire from 4 April 2003 • NAS study on the Future of Supercomputing • Others …

25. How is our task different? • Let’s leverage our SciDAC and other high-end computational science experience to produce a report that: • Identifies “knees in the curve” of scientific knowledge plotted against computer resources in a variety of applications • Describes the implications of important applications for hardware architecture (without delving into a deep discussion of architecture) • Recalibrates how researchers evaluate the merit of a computational attack (hardware and software), with scientific ends in view (not misleading metrics like teraflop/s) • Anticipates algorithmic and software bottlenecks lurking behind new hardware • Describes a well-balanced program to take us past the “knees” (not just hardware, but methods research, software, training, etc.)

26. How is our task different?, cont. • Let’s not major in recommending dollar amounts or retargeting of existing programs [not the charter of our type of committee] • Let’s wax quantitative rather than waxing philosophical [there are plenty of reports that motivate and advocate our transition to an information or knowledge economy in general terms, but nothing readable at a non-expert level that takes on the science case for high-end resources as thoroughly as we are capable of doing]

27. June calendar Tentative dates OFES review meeting DOE Mission Computing meeting NSF meeting (DK) SIAM meeting