France-UK N+N November 2003 1
Grid Projects @ Belfast e-Science Centre Ron Perrott Queens - - PowerPoint PPT Presentation
Grid Projects @ Belfast e-Science Centre Ron Perrott Queens - - PowerPoint PPT Presentation
Grid Projects @ Belfast e-Science Centre Ron Perrott Queens University, Belfast {r.perrott@qub.ac.uk} France-UK N+N November 2003 1 Edinburgh Glasgow DL Newcastle Belfast Manchester Cambridge Oxford RL Hinxton Cardiff London
France-UK N+N November 2003 2
Cambridge Newcastle Edinburgh Oxford Glasgow Manchester Cardiff Soton London Belfast DL RL Hinxton
France-UK N+N November 2003 3
Belfast e-Science Centre Projects GridCast
Television/Radio broadcasting
– Business change, resilience, reliability, cost, customisation, interoperability
RiskGrid
Financial Services
– Business change, performance, cost, resilience, reliability, interoperability
Geddm
High-performance data mining
– Performance, cost, business change, resilience, interoperability
GeneGrid
Bioinformatics
– Performance, cost, business change, interoperability
GridMil
Military infrastructures
– Resilience, reliability, performance, interoperability, agility, cost
France-UK N+N November 2003 4
The GridCast Project
Grid based Broadcast Infrastructures
France-UK N+N November 2003 5
The Grid Scenario: The BBC Nations BBC NI, BBC Scotland and BBC Wales
- BBC Nations provide
customised services in each nation
- Television
programmes are distributed to BBC Nations from BBC Network (London) using dedicated leased ATM circuits.
France-UK N+N November 2003 6
Grid Infrastructure
- Technical
– High-bandwidth network connections inter- connect broadcast locations. – Network bandwidth means geography is less of an issue.
- Organisational
– Less centralised
France-UK N+N November 2003 7
Overview
- To develop a baseline media grid to
support a broadcaster
– Manage distributed collections of stored media – Prototype security and access mechanisms – Integrate processing and technical resources – Integrate with media standards and hardware
- To analyse Quality of Service issues
– Analyse remote content distribution infrastructures – Analyse remote service provision – To analyse reactivity, reliability and resilience issues in a grid-based broadcast infrastructure
France-UK N+N November 2003 8
Characteristics
- Stored media files are Gbytes and increasing
– 1 hour ~ 200 Gbytes; distributes 1 petabyte /year
- Management and distribution is significant
technically
- Metadata – location, timings, artists, storage
formats etc. is an integral part of broadcast structure
- Content is a valuable commodity – access,
modification, copying must be controlled
- High levels of quality required
France-UK N+N November 2003 9
High level view of the Infrastructure
Network Schedule
BBC Network London Controller
BBC NI Belfast
BBC NI Schedule
Controller Transmitter Cable, Satellite, internet BBC Scotland Glasgow Broadcast Output Controller Live Content
BBC Scotland Schedule
BBC Wales Cardiff Controller Live Content Broadcast Output
BBC Wales Schedule Network Schedule
BBC Network London Controller
BBC NI Belfast
BBC NI Schedule
Controller Transmitter Cable, Satellite, internet BBC Scotland Glasgow Broadcast Output Controller Live Content
BBC Scotland Schedule
BBC Wales Cardiff Controller Live Content Broadcast Output
BBC Wales Schedule
BBC Scotland Glasgow Broadcast Output Controller Live Content
BBC Scotland Schedule
BBC Wales Cardiff Controller Live Content Broadcast Output
BBC Wales Schedule
France-UK N+N November 2003 10
Broadcasting Grid Services
Each Broadcast site is defined by its collection of available services
- Control services
- Content services
High Bandwidth IP Network
Local Content Controller
Live Output BBC Northern Ireland
BBC Scotland BBC Wales Controller
Live Output
BBC Network
Network Content
France-UK N+N November 2003 11
A Virtualised Infrastructure
BBC NI BBC Scotland BBC Wales BBC Network
High Bandwidth IP Network
Image Rendering Cluster Video Editing Suite Subtitling Engine Sound Improvement
France-UK N+N November 2003 12
Scenario
- A Network Schedule is defined
– This schedule is the framework for Nation schedules
- Network Schedules are distributed to BBC
Nations
– Usually via email
- BBC Nations formulate their schedule
- A Schedule is Broadcast
– By programming local network and content control automation
France-UK N+N November 2003 13
Model of Broadcast
- Automatic distribution of broadcast
schedules
– Management of schedule archives – Automatic notification
- Content is copied from archives to local
content storage
– Content distribution defined by schedule
France-UK N+N November 2003 14
Broadcast grid issues
- Business change
– A revised organisational model. Services and resources – Each broadcast location gains control….no network schedule.
- Resilience
– Resource sharing and no single programme repository – A BBC Nation can be anywhere!
- Reliability
– Use resources available in other BBC sites or from 3rd party suppliers
- Cost
– Better use of resources and less need for backup resources – Less dependence on particular vendors or suppliers
- Customisation
– Schedule, local resources, local capabilities
- Interoperability
– Business model facilitates sharing with other broadcasters
France-UK N+N November 2003 15
RiskGrid
Grid Financial Services
France-UK N+N November 2003 16
Background
- Financial sector largely cyclical
- Risk assessment calculations
Corporate Intranet Investment Bank (US)
…Traders….
Investment Bank (EU)
…Traders….
Investment Bank (ASIA)
…Traders….
France-UK N+N November 2003 17
Background
- Depends heavily on calculations for competitive
advantage
– Compute intensive
- Large amount of financial derivatives calculations
– Data intensive
- Data-access, bottleneck. 2Gb transactions/day on NYSE
- Improve performance
– Increase accuracy of potential risk in trade – Direct impact on margins – 1% improvement - $$$
France-UK N+N November 2003 18
Architecture
Web/Application Mobile/GPRS Custom RiskGrid Middleware Bus
Database OGSA Adapter
Historical FTSE market data Portfolio databases
Domain 1 OGSA Adapter Domain 2 OGSA Adapter OGSA Adapter Presentation
Publish Bind Publish Bind Publish Bind Utility Computing
France-UK N+N November 2003 19
Issues
- Business Change
– Attempt to give real time risk assessment
- Performance
– Harnessing resources to suit the problem
- Cost
– Use utility and/or unused in-house resources
- Resilience
– Not restricted to locally available resources
- Interoperability
– Provide gateways to other services or service provides
France-UK N+N November 2003 20
GEDDM
Grid Enabled Distributed Data Mining and Conversion of Unstructured data
France-UK N+N November 2003 21
Background
- Fuzzy parallelised data-matching and
transformation engine
- Forensic accounting, banking, anti-
terrorist, crime
- Clusters: PC, Linux, supercomputers
- Large volumes data
France-UK N+N November 2003 22
GEDDM: Business Driver
- Data sources
– numerous structures, formats, locations administrative domains…
- Client
– US County Court: insider trading litigation case
- 45Tb
- Email, pdf, weblogs, RDBMS, Word, files …
- How to process this data to achieve ?
– Meaningful outcomes quickly – Handling multiple formats with common semantic model
France-UK N+N November 2003 23
Issues
- Performance
– Use utility computing to improve performance
- Cost
– Reduce internal need for high performance computing – Reduce the need to provide on-site services
- Business change
– Provide a secure online automated service to companies
- Resilience
– Reduce reliance on internal computing resources
- Interoperability
– Provide mining engine as a service to other services
France-UK N+N November 2003 24
GeneGrid
A Virtual Bioinformatics Laboratory
France-UK N+N November 2003 25
GeneGrid
– Fusion Antibodies Ltd : – Amtec Medical Limited:
- US – NIH, Washington
- Business drivers:
– Develop specialist tissue specific datasets – 3 sites, little collaboration – No dedicated HPC, low bandwidth – Economic advantage (peak demand/supply min) – MicroArray, Seq, Large volumes data…
France-UK N+N November 2003 26
GeneGrid
- Solution
– Grid Service based architecture – Protect confidentiality – Security Model – Genome database integration
- Diagnosis
– Screening protocols aid customised drug targeting – Gene expression profile
- Dataset Mining
– NI stable gene pool & complete patient records – Correlation against various target populations
France-UK N+N November 2003 27
GeneGrid
OGSA Middleware Bus (JMS)
OGSA Gateway OGSA Gateway OGSA Gateway
External Vendor
OGSA Gateway Sequence Data
(Internal) (External)
OGSA Gateway Biological Databases OGSA Gateway Sequence Data OGSA Gateway HPC A OGSA Gateway HPC B