Operation versus generic  apabilities
climate & weather are research

Earth Simulator competitor

$25M/year for 10 yaers

Natural Hazards == Solid Earth

2010 goals are already blessed

application requirements
    observation data

systems requirements -- Ricky Rood
    invest to make systems work
    discipline centered

predictions dominated by observations and
    their assimilation

prediction is initial state problem

Information systems is separate
Instruments is separate

FY02 $21M

ESMF Earth Science Simulation & Modelling Framework

5X data increase

middleware & productivity
DAO
interested in computing futures -- biological computing

evolutionary
revolutionary

NASA is research & development agency

software productivity included

measure of success
need to transfer to operations (National Weather Service etc.)

high end computing is node in information systems
NASA unique features

WANT
Requirements
Technology Roadmaps

Software that computers and networks useful to scientists
Application
Tools
Computer Hardware
Network

So Grids e-ESTO OK

Show Stoppers
Data issues

Software development process
data ingestion management movement
throughput
visualization

What is a Natural Hazard

3 wavelet people!

Lots of Goddard people

Advanced Information Systems

Climate
    2010 100 Tflop sustained
        currently 2.5 gflop best (LOW)
    ESMF "Framework"
    Data management
    more parallelism as bigger problem

Weather
    Management of input data
    2002  64 million observation/day
    2010 5. 10000 million observation/day
    125 factor in computing
    10**7 increase in assimilation
    10 day forecast in 2 hours
    Environmental hazard
        tornado
        air quality
    Disastrous data assimilation problem
    10 day requires global simulation
        need to select just a little amount of
        available data
    Need less but better data!

*********** Second Day ***********************
HEP Data from central accelerator
ESTO data from distributed satellites/sensors
Derived data is distributed anyway

Data sets
    Hierarchical
    Heterogeneous
    dynamic

 

CLIMATE
-------
NSIPP Plan
Want communication technology and fortran compilers

want 400 Teraflop sustained for medium term

pse
    Differentiation
    Data assimilation
    esmf
 
data management
    500 TB per experiment 5 Groups
    
WEATHER
-------
Want Node Compilers as heros can parallelize
so stuff like HPF irrelevant (climate agrees)