Future Directions for NSF Advanced
Computing Infrastructure to support US
Science in 2017-2020
CASC, April 25, 2014
Jon Eisenberg
Director, CSTB
v2
National Academies today
charge to committee
A study committee will examine anticipated priorities and associated tradeoffs for
advanced computing in support of NSF-sponsored science and engineering
research. The committee will consider:
•
•
•
•
•
The contribution of high end computing to U.S. leadership and competiveness in
basic science and engineering and the role that NSF should play in sustaining this
leadership
Expected future national-scale computing needs: high-end requirements, those
arising from the full range of basic science and engineering research supported by
NSF, as well as the computing infrastructure needed to support advances in both
modeling, simulation and data analysis
Complementarities and tradeoffs that arise among investments in supporting
advanced computing ecosystems; software, data, communications
The range of operational models for delivering computational infrastructure, for
basic science and engineering research, and the role of NSF support in these
various models
Expected technical challenges to affordably delivering the capabilities needed for
world-leading scientific and engineering research
reports
interim report (Summer 2014) to identify key issues and discuss
potential options. 2017-2020 time frame
final report (2015) to include a framework for future decision-making
about NSF’s advanced computing strategy and programs. 2017-2022
time frame.
• how to prioritize needs and investments and how to balance
competing demands for cyberinfrastructure investments
• approach: identifying issues, explicating options, and articulating
tradeoffs and general recommendations
NB: no recommendations concerning the level of federal funding for
computing infrastructure
interim report issues
• Trajectory and relevance of large scale simulation’s impact
on foundational advances in science and engineering.
• Scientific research grand challenges that will be
substantially advanced by large scale data analytics and
data mining not currently possible in research
infrastructures.
• Areas for research cyberinfrastructure investments (e.g.
emergent technologies and algorithms, balance between
experimental and “production”, education and workforce
development, community software) required to support
sustained advances in U.S. science
• Challenges and responses by research infrastructures at all
scales (e.g. campus, regional, national; problem focused or
multipurpose) to the items above, identifying those which
can be most positively affected by NSF. These should
encompass economic, cross agency, and international
considerations.
committee
William Gropp, UIUC (co-chair)
Robert Harrison, Stony Brook/Brookhaven (co-chair)
Mark R. Abbott, Oregon State
David Arnett, Univ. of Arizona
Robert Grossman, Univ. of Chicago/Open Data Group
Peter Kogge, Notre Dame
Padma Raghavan, Penn. State
Daniel A. Reed, Univ. of Iowa
Valerie Taylor, Texas A&M
Katherine Yelick, UC Berkeley/LBNL
questions targeted at users
• research needs/opportunities
• advanced computing capabilities, facilities,
requirements
• challenges and suggestions
(draft) questions to inform interim report
research needs/opportunities
What are some of the open problems in your field that require large scale simulation to
solve? Which might lead to fundamental or foundational advances? Why are these
problems not being solved today?
What are some of the open problems in your field that require data intensive computing,
such as large scale data analytics and data mining? Why are these problems not being
solved today?
Are there plans or roadmaps that characterize future computing needs in your field?
(draft) questions to inform interim report
advanced computing capabilities, facilities, requirements
What forms of computing are used in your field? E.g., how does your field make use of
laptop/desktops, research group clusters, department or campus commodity cluster
systems, mid-to large-scale, shared capacity systems such as XSEDE, leadership-class
capability systems such as Blue Waters (NSF) or Mira (DOE), or commercial cloud
services such as Amazon EC2? How would you characterize the importance of access
to each type--required, desirable, or unnecessary? How might these needs change in
the future, and why? Are there kinds of systems that are impractical for the types of
computation that you do or want to do, e.g., because of interconnect performance or
total memory?
With computer hardware and software evolving more rapidly than in the recent past,
what impacts do you see for your field? For example, what role will new hardware such
accelerators (GPUs or Intel Xeon Phi), FPGAs, new memory systems, or new I/O
systems play? Are there barriers to their adoption, such as challenges making
necessary modifications to software?
What software does your field depend on? Who develops and maintains this code, and
how is this work supported
(draft) questions to inform interim report
challenges and suggestions
What are the biggest challenges that your field faces in using computation? Consider
access to systems with sufficient capability and capacity; productivity of environments;
algorithms; workforce; stability of software and hardware; and the ability to use systems
efficiently, including parallelism and scalability.
What investments would have the greatest positive impact on your research field? For
example, this could be more computer systems to increase access, different kinds of
systems with a different balance of capability, support for community software,
development of new algorithms, or a workforce with better training in computational
science.
What other elements of national cyber infrastructure would significantly advance the
pace of discovery or expand participation? For instance, shared file systems, distributed
key-value stores, or standard APIS/services and portals such as managed by XSEDE,
etc.
opportunities for input
• meetings
– May 16
– Teleconferences May 8, 9, 27
• written comments
• report review
• further briefings and discussion to develop final report
 your comments re: study issues, questions
 suggestions for questions, individuals, institutions, programs,
and research areas to consider?
input/suggestions
• Please send suggestions of information sources (people,
reports, etc.) to jeisenbe@nas.edu
• Submit your comments on study topic via National Academies
Current Projects system, http://tiny.cc/inputacistudy
• For more on project, follow www.cstb.org