Transcript Slide 1

Lawrence H. Landweber
National Science Foundation
SC2003
November 20, 2003
[email protected]
• QUESTION 2
Future Cyberinfrastructure
• Built on broadly accessible, highly capable network:
100’s of Tbps backbones;
• Significant distributed and varied computing resources:
100’s of petaflops in aggregate;
• Significant storage capacity: exabyte collections
common;
• Allows wide range of sensors/effectors to be connected:
sensor nets of millions of elements attached;
• Contains a broad variety of intelligent visualization,
search, database, programming and other services that
are fitted to specific disciplines
• The ETF or Extensible Terascale Facility is an early
prototype
Technical Challenges
• How to build the components?
• Networks, processors, storage devices, sensors,
software
• How to shape the technical architecture?
• Pervasive, many cyberinfrastructures, constantly
evolving/changing capabilities
• How to customize CI to particular S&E domains
Shared vs. Special Networks
• Determine application’s networking requirements
including tradeoffs (e.g., cost vs. full or approximate
satisfaction of requirements).
• Determine whether requirements can be satisfied
(modulo tradeoffs) on shared network.
• If not, determine characteristics of network that would
satisfy application's requirements.
• Use an available network with these characteristics.
• If no such network available,
– map characteristics to a network technology and
– implement the target network or some acceptable approximation.
• These are hard problems, not easily answerable for
complex applications and networks.
Research Opportunities
• Develop methodology for mapping applications
to required networks or approximations of
networks.
• Important because not all apps need (or can
financially justify) special networks. Selecting the
target network is not easy.
• There are research opportunities in this area for
collaboration between mathematicians,
computer scientists and scientists.
• QUESTION 5
Next Generation Network Projects
• NSF: ITR Projects
– OptIPuter
– 100 Mbps to 100 Million Homes
• NSF: Extensible Terascale Facility
• National Lambda Rail
100 Mbps to 100 Million Homes
• NSF Funded Research Project (10/03) - $7.5M
• Stanford, Berkeley, CMU, Rice, Fraser Research,
Internet2
• Scope
–
–
–
–
–
Economics
Access
Metropolitan area
Backbone
Protocols
• Requires a redesign of the access, metropolitan and
backbone networks of the Internet
• Applications?
Extensible Terascale Facility
• Resource providers (currently 8)
– Cycles, storage, instruments
• High speed IP interconnect
– Hubs in Chicago and Atlanta
– Links are multiples of 10Gbps
– Reserve dedicated paths for apps
– Management strategies being discussed
• Operational - 2004
2010
• Shared networks for most applications
• Ability to configure special networks to
serve applications
• Dynamically schedulable networks
– Depends on progress in optical switching and
associated management/control
– Eventually in real-time