Planning for CIF21 New Computational Infrastructure

Report
CASC September Meeting
Planning for CIF21 New Computational
Infrastructure:
CDS&E
Software
HPC
Gabrielle Allen, Eduardo Misawa, Manish Parashar
Irene Qualters, Barry Schneider
Office of CyberInfrastructure
September, 2011
1
Work in Progress: Moving from
Recommendations to Plan
• Key elements considered in computational infrastructure:
• CDS&E
• Software
• HPC
• Common themes:
• Operating models for Sustainability- balanced with innovation/agility
• Cross agency and international collaborations
• Effective cross-directorate engagement
• Alignment and overlap with other CIF aspects (e.g. data, access &
security)
NSF-ACCI Task Force on
Grand Challenges:
CYBERSCIENCE AND ENGINEERING
Summary of the TF Recommendations
J. Tinsley Oden
December 9, 2010
Summary and Recommendations:
Computational Methods & Algorithms
General Recommendations
 A broad-based, comprehensive, long-term, and vigorous research
program in advanced computational methods should be established to
overcome the challenges faced in devising, analyzing, and scaling up
new computational methods for critical CS&E problems on advanced
computing systems.
 This program should support multidisciplinary and interdisciplinary
teams that bring together applied mathematicians, computer
scientists, and computational scientists and engineers.
NSF Postdoc Fellowships
CI TraCS: Transformative Computational Science using
Cyberinfrastructure
Image:
W. Ben
ger (LS
U)
.)
olumbiaU
. Turk (C
Image: M
Aim: Support outstanding
postdocs developing innovative
computational and data-enabled
science & engineering
·
Use cyberinfrastructure to make
revolutionary scientific advances
·
Research and develop
cyberinfrastructure to drive innovations
Fellowships awarded
directly to applicants!
$240K over 3 years
Research supplements for fellows
moving to tenure-track faculty
positions
www.nsf.gov/citracs
Deadline: Jan 13th 2012
Applicants must be U.S. citizens, nationals, or legally
admitted permanent resident aliens of the United States
Summary and Recommendations:
High Performance Computing
General Recommendations
 NSF needs to prioritize stable funding of
• a diverse set of classical HPC resources, and,
• innovative computing, networking and data analytics equipment
to support the wide range of needs of the NSF research
community.
 HPC leadership needs development of innovative technologies
including:
• power-aware and application-sensitive architectures,
• new numerical algorithms to efficiently use petascale and
exascale architectures, and
• data flow and data analysis at extreme scale.
Summary & Recommendations:
Software
General Recommendations
It is recommended that NSF:
 Support the creation of reliable, robust science and engineering
applications and data analysis and visualization applications for
Grand Challenges as well as the software development environment
needed to create these applications.
 Provide support for the professional staff needed to create,
maintain, evolve and disseminate the above applications as part of
its grant funding.
 Establish best practices for the release of science and engineering
applications and data as well as the workflows involved in their
creation to ensure the reproducibility of computational results.
Software Infrastructure for Sustained
Innovation (SI2)
http://www.nsf.gov/si2/
Software is Critical to CIF21

SW as the modality for CF21 and
Computational Science in the 21st
Century
CDS&E
 Software is essential to every aspect of
CI – “the glue”
• Drivers, middleware, runtime, programming
systems/tools, applications, …

Software crisis?
 Software complexity is impeding the use of
CI
• Science apps have 103 to 106+ lines, have
bugs
• Developed over decades – long lifecycles
(~35 years)
 Software/systems design/engineering issues
• Emergent rather than by design
 Quality of science in question
CI
SW
Task Force Recommendations






Develop multilevel long-term
program of support
Promote verification, validation,
sustainability and reproducibility
Develop consistent policy on open
source software
Support software through
collaboration of all NSF divisions,
related federal agencies, industry
Utilize Advisory Committees for
input on priorities
These recommendations considered
and acted on through NSF-wide
software working group.
Software Infrastructure for Sustained
Innovations (SI2)

Transform innovations into sustainable
software that is an integral part of
CIF21


Catalyze and nurture multidisciplinary
software as a symbiotic “process” with
ongoing evolution
Create a software ecosystem that scales
from individual or small groups of software
innovators to large hubs of software
excellence
 FY 10: ~$20M
 FY 11: $30M – SSI & SSE
 Under review now
Software Infrastructure for Sustained
Innovations (SI2) – Roadmap

SI2 FY 12 and beyond…
 S2I2 Awards: Small number of S2I2 awards in areas
that are strategically important to CIF21
• Ongoing discussions within the SI2WG
 SSI & SSE Awards: Continued development of core
SW infrastructure through cross-disciplinary SSI and
SSE awards
 Multi-directorate/International Collaborations:
Anticipated formal linkages with NASA, China
(NSFC)and UK (UKRC)
 CI Integration: Activities to integrate SI2 awards
with core CI investments such as XD and DataNet
HPC Task Force Recommendations
1. Develop a sustainable model to provide the academic research
community with access, by 2015–2016, to a rich mix of HPC systems
that:
– deliver sustained performance of 20–100 petaflops on a broad range
of scienceand engineering applications;
– are integrated into a comprehensive national CI environment;
– are supported at national, regional, and/or campus levels.
2. Invest now to prepare for exascale systems that will be available by
2018–2020. NSF should consider the use of co-design partnerships to
provide the HPC systems and data CI needed to enable data-driven
science.
3. Broaden outreach to improve the preparation of HPC researchers and to
engage industry, decision-makers, and new user communities in the use
of HPC as a valuable tool.
4. Establish a continuing process for soliciting community input on plans for
HPC investments.
HPC Strategy in the CIF21 Context
 Transition from TG to XD as the integrating fabric
across HPC resources
 Grow shared services; education, outreach, user support
 Develop XD architecture as a highly adaptable service
 Expand governance model to include new communities
 Separate CI services from hardware
 Common user interface for science not dependent on specific
hardware, geographic location or resource
 HPC resource strategy expands to encompass new
classes and kinds of resources
 XD Managed Resources and Community Provided Resources
 New architectures, clouds, data centers
Align HPC strategy with software, data, networking, campus
bridging and cybersecurity and to support CF21 programs
Blue Waters Update
 ~27 Science/Engineering Projects have been awarded
allocations by NSF under NSF 08-529
 On August 6th, IBM and UIUC terminated their contract for
delivery of the planned Blue Waters system
 Questions regarding the reasons for termination or detail of their
contract should be directed to NCSA or IBM as NSF was not a
party to the contract.
 Numerous public presentations have been given on the intended
configuration and are available online.
 The NSF award to UIUC is in the form of a Cooperative
Agreement
 UIUC has requested a change to the PEP– this change will be
peer reviewed by external panel in Sept.
FY11 HPC Resource Solicitation
 Timeline – NSF 11-511




December, 2010 Solicitation Issued
March 7, 2011 Proposals due
September, 2011, One award of up to $30M anticipated
January, 2013 Initial availability in XD
 Includes Community Feedback from ACCI HPC taskforce
• Option of an additional 4 year renewal award for
upgrade/replacement
• Both computational and data intensive
 Voluntary Cost Sharing Prohibited per new GPG
 O&M award (up to $6M/annum) separate
 Introduction of “Innovative Capability” requirement
 Optional $5M/$30M for this capability
HPC Strategy – 3 streams
 Transition from TG to XD as the integrating fabric
across HPC resources
 Revisit HPC resource strategy to consider new classes
and kinds of resources
 XD Managed Resources and Community Provided Resources
 New and diverse architectures, clouds, data centers
 Promote inclusion of all scientific domains
 Promote flexible availability at all scales
 Adjust /replace existing solicitation to match strategy
 Align with other CIF21 elements (e.g. software, data
and campus bridging) thru CIF21 WG

similar documents