Presenter: Sabine Roller [roller@hlrs.de]Coordinator: Hubert Hérenger [herenger@hlrs.de]

High Performance Computing Center Stuttgart (HLRS)

Business Experiment 06 - Groundwater Modelling

Providing better forecast possibilities in groundwater modelling using optimization and grid technologies

Business Experiments in GRID2

Project Data Sheet

• Type of project: Integrated Project• Project coordinator: Mr. Santi Ristol

santi.ristol@atosorigin.com (ATOS ORIGIN)

• Project start date: 1st June 2006• Duration: 42 months (Nov 2008)• Budget: 24.7 M Euros • Max EC contribution: 15.7 M Euros (63%)

• Consortium: 75 partners • Effort: 2713 PM (226 PY,65 P,360.000h)

The mission of BEINGRID is to Exploit European Grid middleware by creating a toolset repository of Grid services from across the Grid research domain and to use these services to deliver a set of successful business experiments that stimulate the early adoption of Grid technologies across the European Union.

Business Experiments in GRID3

Business Experiment 06

BE 06

Service

provid

er & B

E leaderApplication provider & consultant

Inte

gra

tor

Be

ne

fits

Be

ne

fits

End user

End user

End user

• BE06 Goals:– Large scale multidisciplinary

compute grid– Improved prediction via Grid

technologies– Business models for license

management– Designing, implementing and

validating solution

• BE Details:– Duration: 20 months– Economic sector: SMEs

providing services for groundwater modelling

– Grid middleware: Globus Toolkit 4, Unicore/GS

Business Experiments in GRID4

Decision problems related to groundwater

• important resource for water supply• important component of aquatic environment

endangered world-wide by

• depletion due to overexploitation• pollution by industry, land-use, hazardous wastes, etc.

Problem:

Large time gap between cause and consequences of system changes!

Business Experiments in GRID5

Decision problems related to groundwater

Example: Pollution within the catchment areas of water works

• lack of information on exact position of contamination sources • limited knowledge on effects of pollution (spatial-temporal distribution)

Many effects are notified first, if large quantities of water are polluted and remediation measures become expensive and cumbersome.

-> there is an urgent need to gain information on the potential risk from polluted sites as early as possible, to forecast possible risk and to design remediation strategies!

Business Experiments in GRID6

Groundwater models

Complex groundwater models

• finite element models/simulation (FE-models)• 3D• Transient (non stationary)• coupled flow and mass transport • high computational effort

– some 100.000 elements– some 1.000 time steps

Standard method for decision support

scenario analysis

Business Experiments in GRID7

Sensitivity analysis

The sensitivity analysis is carried out to• analyze uncertain parameters by• several parameter variations and • therefore to estimate the model uncertainty

Current problems by performing sensitivity analysis• no (semi-)automatic implementation available• thus it requires a lot of manual work• very long runtimes due to a large number of simulations

sensitivity analysis is often disregarded during modeling

Business Experiments in GRID8

Model based optimization

Background of model based optimization• non-linear correlation of flow- and transport processes• therefore complex analysis are required to get the optimum system

configuration• optimization algorithms are powerful instruments to get optimum model

setting with respect to constraints

Disadvantages of optimization algorithms are• hugh amount of time and computational power due to a very large number

of model runs• often model and/or optimization software must be adapted manually

Off1

On1

Off3

Pumping station

Switch level

Groundwater level

Off2

On2

On3 Off4

On4

Constraint

Business Experiments in GRID9

Customer requirements

Groundwater models are often used by• small engineering companies,• public authorities,• water suppliers and• without own informatics section

Model design is often a service delivery

Model usage must require non or at most a minimum of informatics knowledge

Business Experiments in GRID10

Customer requirements

So model calibration, sensitivity analysis and optimization can be provided

• as supply of services or • as a “toolkit” providing

– groundwater relevant “easy to use” graphical interface,– applicable runtime behavior,– evaluation methods and– without need of informatics or programming skills

Business Experiments in GRID11

The status quo: distributed optimization

FEFLOW

2. Simulation run

1.1. Start simulationStart simulation

3. Return simulationReturn simulationresultsresults In

terf

ace

Managem

ent

Middleware

Cluster of workstations

Optimization

Business Experiments in GRID12

The status quo: distributed optimization

The distributed optimization• provides an essential speedup of optimization and• so enables the optimization of groundwater problems even in complex

cases• nevertheless, mathematical correct optimization of complex problems

(transient, mass transport) are still hard to hand • but: optimization (even with simplified assumptions) provides a better

screening of rational scenarios• for water quantity problems (e.g. optimization of groundwater lowering) the

method is practicable already

Using the present distributed optimization requires modeling and development skills, so optimization requires at least a modeler and a computer scientist

Business Experiments in GRID13

The goals: providing a grid toolkit

The goal of the “grid toolkit” is

• to provide a “FEFLOW ® grid toolkit” including– a FEFLOW ® grid service to simulate FE-models ,– grid services to enable distributed model calibration, sensitivity

analysis and/or optimization– graphical “easy to use” user interfaces

• to integrate the toolkit into a grid framework (e.g. GT4, Unicore/GS)• design of a portal solution to access the services (e.g.

GridSphere/GridPortlets) and/or • to access grid resources by the actual FEFLOW ® application and/or

standalone client.

Business Experiments in GRID14

The goals: providing a grid toolkit

To succeed with a grid toolkit basic features have to be:

• “easy to use” user interfaces,• a secure and reliable environment,• access to new resources (thus performance gain) without specific

knowledge,• problem oriented predefined scenarios (to be easily adapted to different

models) and• a efficient business plan (accounting, billing).

Business Experiments in GRID15

The goals: providing suitable business models

A key aspect of BEinGRID is bringing new grid technologies out of the lab into the real world.

• Currently loads of technologies developed in different EU projects wait for adoption

• Obstacles to adoption are:– smaller technological gaps resulting from many projects being very

domain specifc (these will be closed in the BEs and other workpackages)– lacking business models

• To succeed with a grid solution appropriate business models are required:– suitable for service consumers, e.g. end users from different sectors

(SMEs, governmental) …– suitable for service providers and ISVs, e.g. WASY

Thank You.