IEEE eScience: Final Keynote Edward Seidel, Director, Office of Cyberinfrastructure, NSF (since the summer, I guess)

he’s a physicist and a computer scientist, formerly of LSU

how he got here – started in HEP then moved to general relativity-
- a series of problems that require supercomputers, computer scientists, high speed nets, grids, visualization
black hold perturbation theory – very hard, try supercomputers
black hole collisions
neutron star collisions

themes –
costs a lot, requires collaboration, idea to reuse tools..
coastal modeling – an example for cyberinfrastructure
another example LHC – quantity of data

data driven era of science

NSF vision (see Atkins report 2003 – Cyberinfrastructure Vision for the 21st century)
1 virtual organizations for distributed communities (large-scale teams to solve complex problems)
2. hpc
3. data visualization/interaction
4. learning and workforce

Data Net
- developing communities and tools to solve complex problems –example climate change : overlaying chemistry, environmental, etc.,

Virtual organizations for distributed communities

learning and workforce development
- need computational science (not cs but broader?) programs

Teragrid
- track 2 (Texas, Tenn, Pittsburgh… another under review)
- track 1 petascale university of Illinois – but will only serve a small number of scientists

next generation – xd – xtreme digital (still looking at this)
- innovative ways to support digital services, has explicit visualization component

open science grid and loosely coupled science grids
- integrate national or international needs into campus needs/support
- some of this is aimed at the LHC, but there are other efforts that can use it, too

but what about software?
- no real program at NSF to build the software to take advantage of these nets
applications to take advantage of this
virtual organizations to work out how to do all of this

Blue Waters
IBM Power 7 based system
online 2011
1 petaflop sustaind performance on real applications
>200k cores

idea is not a whole bunch of jobs – but a few jobs that need this kind of complexity
coupled dynamic ensemble simulations, real-time simulations – real policy issues like scheduling

will keep going with the international programs

translation to programs – get bandwidth to the center but not to the labs where its needed
federated id management

what next?
pick up and do remainder of things in vision
need end-to-end integration
need to support a computational science community

he comes up with the third pillar, too, experiment, theory computational science

Labels: IEEEeScience08

  ¶ 2:14 PM| |cites (technorati) |