World's Largest Working Computing Grid

fenimor writes "UK particle physicists claim that they will demonstrate the world's largest, working computing Grid with over 6,000 computers at 78 sites internationally. The Large Hadron Collider Computing Grid is built to deal with 15 Petabytes of data each year from the Large Hadron Collider (LHC), currently under construction at CERN in Geneva. 'This is a great achievement for particle physics and for e-Science,' says Professor Tony Doyle, leader of GridPP. 'Our next aim is to scale up the computing power available by a factor of ten'."

Comments

Computing power

[nemexi]

Does anybody know facts about the computing power of the grid? How many teraflops will it be able to achieve?

Grid vs. LHC@Home?

[Anonymous Coward]

What's the point of the Grid thingy if they've also setup this?

http://lhcathome.cern.ch/

Coordination

[erick99]

I wonder how they are coordinating the use of all of those computers? The article doesn't say that they will be exclusively for this project and, if they are not, then that is some task to have them all online and not otherwise busy. They must have some damned serious storage vaults as well if they are generating 15 Petabytes a year of data, which doesn't include the output from processing. Still, it must be something to have all of the "horsepower" at your command.

Cheers,

Erick

Re:Coordination

[Anonymous Coward]

I wonder how they are coordinating the use of all of those computers?

Carefully. Well, with some very complex schedulers and batch systems and LDAP directories and SQL databases and and bits from the Globus project and lots of other scripts and random crap. It's kindof a miracle it all works (when it works...), a bit like the Internet itself really.

Re:Coordination

[steve_l]

I have access to some of the machines; we donate idle systems to the project in exchange for low cost (read free) access to the superjanet network. When they arent doing UK NeSC grid stuff I can bring up vmware images of whatever distro I feel like, run whatever stuff we need -in my case usually distributed testing of distributed software.

That is how the grid works -it uses spare cycles on machines in the network. Unlike Seti@home, they are very fussy about bandwidth; you need a serious link to play. Most of the tier-1 sites are UK academic sites -rutherford labs, oxford, the london universities, etc. Tier-2 sites are us industrial sites with machines we let them play on.

Like you say, data storage is a big issue, and something the grid needs to work on. A lot of grid forum work is on data.

Incidentally, when the LHC comes on line, then the serious data starts to collect. Any collisions -events, they call them- generate vast amounts of data. There is some logic in the system to immediately split dull events from potentially interesting ones, but those interesting ones happen often enough you need to buffer it all up, then do more rigorous analysis to see if it is something new or something known about. The uk grid will be used for analysis of stuff the front end thinks could be interesting. Right now it can test the data generated by simulations (people can get phds writing good event simulations), so they can verify that the analysis code works.

One of the great fear in high energy physics is that the filter and detection logic is buggy and that nobel-grade events are missed because the data is simply too bizarre for the analysis code, and so gets missed out,..

Largest?

[anethema]

I am not sure how they define largest...

Are these 6000 super computers? Or just other computers?

Distributed.net had around 330 thousand participants on the latest completed rc5 key.They had 15 thousand active on the last day of the challenge.

I would say this is much larger in computer numbers, but since they dont mention almost any usefull information in the article, I'm not sure if more computer power would be in the d.net.

However the line: By 2007, this Grid will have the equivalent of 100,000 of today's fastest computers working together to produce a 'virtual supercomputer', which can be expanded and developed as needed

So right now it isnt even 100 thousand computers, maybe not even close, so the computing power might be similar. (assuming 15 thousand active computers on d.net)

Either way, right now i highly doubt its the largest ;)

One tiny problem ...

[Anonymous Coward]

One of my buddies was an early numerical modeller. If I learned one thing from him it was that all the computer power in the world was no use if your model was even slightly defective. The models tended to 'blow up'. Imagine a hundred foot wall of water moving majestically down the estuary.

Typical of stories about these giant computers, they don't really describe the problems they intend to solve. In a way, that is the more interesting story. Mind you, that story is much harder to tell if you want your audience to understand it.

Re:Seamlessly? I doubt it. Latency is a big proble

[tkittel]

True, and this makes it difficult for people who want to calculate protein folding or predict next weeks weather. But for particle physics computations we hardly need any communication between nodes at all. Rather, we need something simulated a huge amount of times (as in, "simulate this proton-proton collision 10 billion times") or "apply this fancy pattern recognition algorithm to each of these billions of events we took this week". Particle physics computations are to a large extent parallel in nature from the beginning.

The grid related problems faced in particle physics are of another nature, such as ensuring that the data is copied around the various grid facilities as needed and of ensuring that even if a given node fails to execute its job for some reason it is rerun elsewhere automatically - that sort of thing.