Meteo-GRID: World-wide Local Weather Forecasts by GRID Computing

Meteo-GRID:World-wide Local Weather Forecasts

by GRID Computing

Deutscher WetterdienstPO Box 10 04 65D - 63004 Offenbach am MainGermany

e-mail: claus-juergen.lenz@dwd.dehttp://www.dwd.de

C.-J. Lenz, D. Majewski, G.-R. Hoffmann

Claus-Jürgen Lenz, Detlev Majewski

Contents:

- Introduction to EUROGRID and Meteo-GRID

- Detailed Description of Meteo-GRID, computational requirements, status of work

- Demonstration example

Gesellschaft für Parallele Anwendungen und Systeme mbH

Pallas GmbHHermülheimer Straße 10D-50321 Brühl, Germany

info@pallas.comhttp://www.pallas.com

www.eurogrid.org

Application Testbed for European GRID computing

Volume: 33 person years, 2 Million Euro fundingby European Commission Grant No. IST-1999-20247,Funding time: Nov. 2000 - Oct. 2003

EUROGRID Vision

Build a European GRID infrastructure

that gives users

a seamless, secure access to

High Performance Computing resources

and that advances computational science

in Europe

EUROGRID Goals

- Integrate resources of leading European HPC centres

into a European HPC GRID

- Develop new software components for GRID computing

- Demonstrate the Application Service Provider (ASP)

model for HPC access (‘HPC portal’) for different

applications

- Contribute to the international GRID development

Structure of the Work

• Application GRIDs:application-specific interfaces, evaluation of GRID solutions– Bio-GRID– Meteo-GRID – CAE-GRID

• HPC GRID Infrastructure:connect HPC centers using UNICORE technology

• Development and integration of new software components

• Dissemination and exploitation

European Testbed for GRID ApplicationsBio-GRID

• Operate a GRID for biomolecular simulations

• Develop interfaces to existing biological andchemical codes

Meteo-GRID

• Develop a relocatable version of DWD‘s weather prediction model

• Goal: ‘Weather prediction-on-demand‘ as an ASP solution

CAE-GRID

• Coupled simulations of aircrafts

• HPC portals for EADS engineers and for engineers at Daimler-Chrysler and partners

• Develop GRID technology for computing cost estimates and billing

HPC Research GRID

• Demonstrate a European HPC GRID testbed

• Develop new GRID applications

• Enable sharing of competence and know-how

• Agree on security standards, certification, access policies, ...

Technology Development• Build on the functionality of UNICORE

• Extend UNICORE to provide the middleware necessary for the Domain specific GRIDs

- Efficient data transfer

- Resource brokerage

- ASP services

- Application coupling

- Interactive access

EUROGRID Partners

HPC Centres

• CSCS Manno (CH)

• FZ Jülich (D)

• ICM Warsaw (PL)

• IDRIS Paris (F)

• Univ Bergen (N)

• Univ Manchester (UK)

Users

• Deutscher Wetterdienst

• EADS• T-Systems

(Assistant Partner)

Integration• Pallas (Project Coordinator)

• Fecit (Assistant Partner)

Goal of Meteo-GRID

To provide

high-resolution short range weather forecasts

with the

relocatable nonhydrostatic “Lokal-Modell” (LM)

of DWD

for any desired region in the world

Meteo-GRID

• Develop a relocatable version of DWD‘s weather prediction model

• ‘Weather prediction-on-demand‘ as an ASP solution

Meteo-GRID: Meteorological Portal

Hoffmann (DWD)

Meteo-GRID: Potential Users

• Use by other meteorological services

• Use by weather service providers

- commercial application

• Use by individuals via Internet

- weather forecast on demand

• Use by individuals via Mobile Telephones

- WAP services

Hoffmann (DWD)

What´s special about Meteo-GRID ? (1)

- Real-time weather forecasting is a time-critical task,

a 48-h forecast must be completed in less than

60 minutes

- LM is a large MPP code of about 100.000 lines of code,

Fortran95, MPI for message passing

- Weather forecasting is computationally expensive

~ 4000 Flop/grid point and time step

~ 15 Tflop for a 48-h forecast (160 x 160 x 35 grid points,

grid resolution ~ 7 km)

~ 3000 sec at a sustained speed of 5 Gflop/s

CPU requirements of LM

What´s special about Meteo-GRID ? (2)

- Weather forecasting requires high band width for data

transfer

Forecast data (at hourly intervals): (48+1) x 20 Mbyte = 1 GByte

Transfer in less than 1 hour: 2.4 Mbit/sec

- “Weather” has large social and economic impact

worldwide (storms, floodings, snow, freezing rain ...)

Damages following cyclone Lothar in southwestern Germany (26 Dec 1999)

Flood at Vistula river, summer 2001

Coastal storm at Hamburg and at the North Sea Blizzard in New York

Tasks of Meteo-GRID (1)

Selection of

- model domain,

- grid resolution,

- forecast date,

- forecast range and

- forecast products

using a Graphical User Interface (GUI)

Meteo-GRID GUI (1)

Nellari, Ballabio (CSCS Manno)

Meteo-GRID GUI (2)

Nellari, Ballabio (CSCS Manno)

Tasks of Meteo-GRID (2)

Derivation of

topographical data

for the selected model domain from high-resolution

(1 km x 1 km) data sets (GIS) at DWD

Examples for TOPO applications (1)

Examples for TOPO applications (2)

water

peat

clay

loamy clay

loam

loamy sand

sand

rock, concrete

ice, glacier

undefined

Examples for TOPO applications (3)

Tasks of Meteo-GRID (3)

Derivation of

- an initial data set and

- lateral boundary data sets

for LM from data of the global model GME of DWD

(Oracle data base)

GME model grid and LM domain

Tasks of Meteo-GRID (4)

- LM forecast run is performed on any supercomputer

available in EUROGRID using UNICORE technology

- Forecast data (GRIB code) are returned to the user

via UNICORE and the Internet

OR . . .

Tasks of Meteo-GRID (5)

OR . . .

- Visualization of LM forecasts ( 1 to 5 dimensional

graphics) on the HP Computer and subsequent

- Return of image files to the user via UNICORE

and the Internet

- Verification and validation of LM forecasts for any

region worldwide

Information and data flow in Meteo-GRID (1)

1.1. Set up of LM-domain Set up of LM-domain

User

Global topographicaldata set (GIS),~ 7 GByte

Topographical data set(1 - 5 MByte)

DWDDWDDWDDWD

GUI:GUI:Selection of- domain corners- grid resolution- forecast date- forecast range- forecast products

Calculation at DWD onSGI Origin O 2000 orIBM RS/6000 SP (5 - 30 min wallclock time)

Information and data flow in Meteo-GRID (2)

2. Define forecast date and range

GME data base (Oracle)

Hourly initial and lateral boundarydata sets on GME grid (~ 50 MByte)

User DWDDWDDWDDWD

Extraction of GME results covering the LM domain at DWD (SGI Origin O 2000 or IBM RS/6000 SP)

Information and data flow in Meteo-GRID (3)

3.3. Perform LM-forecast on EUROGRID HPC and send forecast data to userPerform LM-forecast on EUROGRID HPC and send forecast data to user

User DWDDWDDWDDWD

Topographical data set

Initial and lateral boundarydata sets on GME grid

HPCGME2LM

interpolation of GME results to LM grid

LMcalculation of weather forecast

1 - 5 MByte

1 - 20 GByteLM-forecast datavisualisation

~50 MByte

Initial and hourly lateral boundarydata sets on LM grid (1 - 20 GByte)

For more Information . . .

about DWD: http://www.dwd.de

about Pallas GmbH: http://www.pallas.com

about UNICORE: http://www.unicore.de

http://www.fz-juelich.de/unicoreplus/index.html

about EUROGRID: http://www.eurogrid.org