TeraGrid n US based, funded by the NSF n High speed network, combines clusters into one Super-Cluster! n Doesn’t just combine computation… combines storage,

TeraGrid

US based, funded by the NSF High speed network, combines clusters

into one Super-Cluster! Doesn’t just combine computation…

combines storage, sensors, visualisation, all computing resources!

Enables better science

TeraGrid as a Distributed System

Computing resources at different sites TeraGrid to be used as a single system Systems built from individual nodes

Not the only Grid…

Many grids worldwide Examples: EUGrid, I-Grid Individual clusters & supercomputers in

various places, commercial and academic

What can we use Grids for?

Hard & heavy research Applications:

– Geography, weather forecasting– Economics, financial predictions in markets– Physics, simulations, theory verification– Chemistry & Biology, AIDS/Cancer– Raytracing, Encoding Audio/Video, more!

Image/Video Rendering

Toy Story made on a distributed system, a cluster!

117 Sun SparcStations Render farm, a load of machines

rendering different frames of video 114,000 frames, 77 minute movie One computer would have taken 43

years of nonstop computing

Parallel Video Encoding

MPEG

Immersive Environments

Video Walls! Visualisation stuff

Analysis of Video

Shot & Scene detection in Fischlar Slow computation process, goes

through every frame of video to analyse Our beowulf does it 5 times faster than

a large enterprise Sun server, worth 10G!

Search engines

Google, pigeon rank Searching wide database, ideal for

deployment on a cluster *was* driven by 4500 node Linux

Cluster

Breaking Security - Computing

RC5, DES, & distributed.net 2^64 (18,446,744,073,709,551,616)

keys Teams, donated CPU cycles

Other “proper” research

Molecular modelling, drug design Study of viruses, virus propagation AIDS/Cancer research Weather forecasting, climate prediction High Energy Physics Simulation, earth, economics, financial Artificial Intelligence

Other stuff in Distributed Systems

Distributed Software Objects, CORBA Distribued Information = WWW Peer to Peer – Napster and mates

What issues are there?

Organisation Resource Management, Scheduling Optimisation Security Balancing load Handling errors

Linux Cluster

How do they do it?

Split up problems, requires specific analysis for different problems though :-/

Architecture of a cluster, software Parallelising Compilers, why not do it

automatically?

Distributed Systems

Loose definition - collaborative computing

Enables all applications in all areas, however mostly for large stuff

Pushes back the frontiers of what we think we’re able to solve

Better use of our scarce resources!