Geodise: Grid Enabled Design Optimisation and Design Search Applications and Testbeds Working Group

1
Geodise Grid Enabled Design Optimisation and
Design SearchApplications and Testbeds Working
Group Workshop14th MarchGGF10 2004,
BerlinProf Simon CoxTechnical
Director,Southampton Regional e-Science
Centre,Computational Engineering Design
GroupSchool of Engineering SciencesUniversity
of Southampton
2
Grid Enabled Optimisation and Design Search for
Engineering (GEODISE)Southampton, Oxford and
Manchester

• Simon Cox- Grid/ W3C Technologies and High
  Performance Computing
• Global Grid Forum Apps Working Group
• Andy Keane- Director of Rolls Royce/ BAE Systems
  University Technology Partnership in Design
  Search and Optimisation
• Mike Giles- Director of Rolls Royce University
  Technology Centre for Computational Fluid
  Dynamics
• Carole Goble- Ontologies and DARPA Agent Markup
  Language (DAML) / Ontology Inference Language
  (OIL)
• Nigel Shadbolt- Director of Advanced Knowledge
  Technologies (AKT) IRC

• BAE SYSTEMS- Engineering
• Rolls-Royce- Engineering
• Fluent- Computational Fluid Dynamics
• Microsoft- Software/ Web Services
• Intel- Hardware
• Compusys- Systems Integration
• Epistemics- Knowledge Technologies
• Condor- Grid Middleware

3
The GEODISE Team ...

• Richard Boardman
• Sergio Campobasso
• Liming Chen
• Mike Chrystall
• Trevor Cooper-Chadwick
• Simon Cox
• Mihai Duta
• Clive Emberey
• Hakki Eres
• Matt Fairman
• Mike Giles
• Carole Goble
• Ian Hartney
• Tracey Hunt
• Zhuoan Jiao

• Andy Keane
• Marc Molinari
• Graeme Pound
• Colin Puleston
• Nicola Reader
• Angus Roberts
• Mark Scott
• Nigel Shadbolt
• Wenbin Song
• Paul Smart
• Barry Tao
• Jasmin Wason
• Fenglian Xu
• Gang Luke Xue

4
Design Challenges

• Modern engineering firms are global and
  distributed
• Not just a problem of using HPC

How to ?
CAD and analysis tools, user interfaces, PSEs,
and Visualization
improve design environments cope with legacy
code / systems
Optimisation methods
produce optimized designs
Management of distributed compute and data
resources
integrate large-scale systems in a
flexible way
Data archives (e.g. design/ system usage)
archive and re-use design history
Knowledge repositories knowledge capture and
reuse tools.
capture and re-use knowledge
5
GEODISE
Geodise will provide grid-based seamless access
to an intelligent knowledge repository, a
state-of-the-art collection of optimisation and
search tools, industrial strength analysis codes,
and distributed computing data resources
6
Coming up next

• Application
• Compute
• Data
• Workflow construction
• Knowledge

7
Application
8
Design
9
Computational Fluid Dynamics
10
Engine Nacelle Optimisation (problem definition)
Assumption Noise radiated to ground reduces
with increasing scarf angle
Objective function Total Pressure
Recovery (pt2/pt1) Design variables Scarf Angle
(degrees) Axial Offset (mm)
11
Engine Nacelle Optimisation (mesh generation)
Table 1 Relations determining the edge node
spacing based on geometric information
12
Engine Nacelle Optimisation (3D) (some results)

• Typical unstructured mesh used in the problem
  (left)
• Response surface model built for two design
  variables (right)
• The effect of other geometry parameters need to
  be investigated

13
Design of Experiment Response Surface Modelling
14
Defining the Objective Function

CAD geometry
Design Variables x1 0.5, x2 0.25
Meshing
CFD analysis
Objective function y 42
Post-processing
15
Photonic Device Modelling
pitch300nm
Bridge waveguide structure courtesy of Martin
Charlton, Southampton Microelectronics Research
Group.
REAL-THING (photo)
PERIODICALLY TILED UNIT-CELLS
UNIT-CELL
16
12-fold Symmetric Quasicrystals

• Based on tiling of dodecagons composed of squares
  and equilateral triangles
• Possesses 12 fold rotational symmetry
• Leads to a highly homogeneous band gap

17
Quasicrystal Results
FDTD Results
Experimental Results
FEM Results (Density of States)
18
CEM Simulation Results
CompResource.1 CompResource.2 CompResource.3 CompR
esource.4 CompResource.1
50 radii
50 frequencies
Photonic Crystal Response Surface / Photonic Band
Gap Map
19
Photonic Crystal - Optimisation
User
CAD Model Parameters
CAD Model Parameters
Location, size density of holes determine the
optical bandgap.
Discretization
On the Grid
Numerical Solver
Intermediate Result
The aim of this design optimisation process is to
find a configuration which maximises this bandgap
and minimizes energy loss.
Optimised Design
User
User Molinari CAD Model A Parameter1
Rad Parameter2 Pos Optimise BGap Computed X Iter
Log File
20
Compute
21
Grid-Enabled Scripting Environment

• Motivations
• Flexible, transparent access to computational
  resources
• Easy to use for engineers
• Our Approach
• Matlab chosen as the hosting environment
• Extend the users existing PSE
• High-level functionality
• Quick application development
• Computational resources exposed in the form of
  Matlab functions
• Job submission to Globus server using Java Cog
• Job submission to Condor pool via Web services
  interface
• Integration of CAD, Mesh generation, and Fluent
  solver via the use of intermediate data format,
  often standard-based, or package-neutral
• Hybrid search strategies to make the best use of
  different search methods

22
Geodise Architecture
Integration Scripting
Knowledge Services
Matlab (or Jython )
Intelligent Support
Java / C
Interface
Web Service Grid Service
Java / C/ .NET
Building Blocks
.EXE/ Fortran/ Matlab Code
23
Security

• Certificate based
• UK e-Science CA
• End-to-End
• WS-Security- public-private key signed for
  recipient
• DIME for large application files

24
Grid-Enabled Toolkits in Matlab
Pound, G.E., Eres, M.H., Wason, J.L., Jiao, Z.,
Cox, S.J., and Keane, A. J., A Grid enabled
Problem Solving Environment (PSE) for Design
Optimisation within Matlab, 17th International
Parallel and Distributed Processing Symposium
(IPDPS 2003) 22-26 April 2003, Nice, France, 2003
25
Geodise Matlab Computation Toolkit for Condor
Condor Pool
26
Scripting the optimisation workflow within Matlab
27
Data
28
Data Challengesbring database tools to
engineers

• Large quantities of data generated at different
  locations with different characteristics.
• Engineering data is traditionally stored in flat
  files with little descriptive metadata
• Hard to search and share
• Hard to re-use
• Our focus is to leverage existing database tools
  not commonly used in engineering applications,
  and
• provide them in an environment familiar to
  engineers

29
Metadata-Assisted Data Modelling and Archive
Toolkit

• Metadata
• Data about data
• Assisting data archive and query
• Grid-enabled data access
• Data, including files and Matlab structures are
  archived in database with metadata attached
• Queries can be carried out on the metadata
• Access to data is specified by the owner of the
  data
• Uniquely named data elements using UUID
• Implementation of hierarchy data structures
  using cross reference
• Authentication required to archive and query the
  data

Wason, J., Molinari, M., Jiao, Z., and Cox, S.J.,
Delivering Data Management for Engineers on the
Grid, EuroPar2003.
30
Database Architecture
31
Application Specific Data Structure
32
Client Tools (1)

• Archive data - gd_archive
• Store files/ structures into archive with some
  metadata.
• metadata.model pgb_design
• metadata.result.bandgap 20
• fileID gd_archive(C\input.dat, metadata)
• var.a 1.4, 5.32, 4.98
• structID gd_archive(var, metadata)
• Group data - gd_datagroup, gd_datagroupadd
• Logically group together related data.
• groupID gd_datagroup (my datagroup,
  group_metadata)
• gd_datagroupadd (groupID, fileID)

33
Client Tools (2)

• Query data - gd_query
• Query archive from script or GUI.
• gd_query (file.userID me
• result.bandgap lt 40,
• file.)
• Retrieve data - gd_retrieve
• Retrieve archived data to local machine.
• gd_retrieve (fileID, E\files\control.dat)
• var gd_retrieve (structID)

34
Workflow
35
Workflow Construction, Validation, Monitoring
and Steering
lt?xml version"1.0" encoding"UTF-8"?gt ltFunction
sgt ltTestfunctionsgt ltrand
type"Instance"gt ltinputsgt
ltinput1 type"int" name"n" value" 5 "/gt
lt/inputsgt ltoutputsgt
ltoutput type"matrix" name"A" value""/gt
lt/outputsgt lt/randgt
lt/Testfunctionsgt lt/Functionsgt
Submit to Matlab
DB
36
Conclusions

• Delivered production quality environment
• Legacy Integration
• Scripting Powerful
• Compute
• Data

37
Questions