By Jo Best, 5 January 2005 16:50
Grid computing. Sounds like something out of the Matrix. What is it really?
It's a sort of Lego-style approach to computing - usually supercomputing.
Go on.
Grids connect up computers - usually not in the same building or even the same country - with the idea of joining up power and resources of disparate machines. The term 'grid', coined by Dr Ian Foster, refers to the idea of electricity grids - the theory behind both computer grids and electric grids being that you get power whenever you need it, with power going in from any number of disparate sources.
That's it, then? That's a grid?
Well, yes and no. Grids are still loosely defined and not even those working on them necessarily agree on what they are. For example, some business grids are used to connect up a load of PCs in one building. Some grid researchers consider that an intra-grid, not a grid proper, for example.
Grid - that's all about finding aliens, isn't it?
I suppose you mean SETI@Home - where people around the world contribute their spare computing power to analyse radio telescope signals in the hope of finding some extra terrestrial life. Some would say that's a grid though its creators say it isn't. Either way, there is more to grid than aliens.
Like what?
A lot of the grid projects up and running at the moment are in academic or research institutions. The biggest grid working at the moment, part of the Large Hadron Collider, has over 6,000 machines in 78 countries, all working away on problems of particle physics. Another grid is being used by the Met Office to predict climate change and the EU has its own grid project, the Enabling Grids for E-Science in Europe project.
So it's just boffins that use grids then?
Not at all - it's getting a lot of attention in the business world too. Financial institutions and Wall Street staples in particular have been early adopters of the technology. Several of the major vendors involved in flogging grid - including HP, IBM and Sun - have been trying to tempt more bankers into using the technology with 'pay as you go' style offerings, the idea being that the City boys can rent a bit of extra computing power from the vendors when they need it.
As well as the number crunchers, grid has seen some interest in the retail and transport sector.
Right. All sounds a bit niche to me - when will it be mainstream?
Analysts seem to favour the 'next few years' as a likely target, with one estimate putting the grid market as being worth some $12bn by 2007 - but, like most new technologies, it'll surface first in non-mission critical applications before sceptical CIOs will pluck up the courage to get more friendly with grid.
So what's the hold-up then? If it's such a great idea, why is no one doing it?
Like most new technologies, standards are one of the main problems. And with grid intending to connect up any number of machines potentially running a huge range of software between them, middleware - and therefore standards - is crucial.
Given that standards are still being defined in various areas such as apps and data management, that more than one body, including OASIS and the Global Grid Forum, is working on defining them and that a vendor-led coalition is not ruling out the possibility of developing its own standards, some would-be grid users are waiting for the area to mature a little before jumping in.
Currently, however, the Globus Toolkit - put together by the founding fathers of grid, Dr Ian Foster, Steve Tuecke and Carl Kesselman - is the de facto standard for the technology.
So where's grid going?
Some people have speculated that one possible result could be that grid could eventually grow either to replace or enhance the internet, with spare computing power distributed around the world when it's needed, where it's needed.
That's beautiful.
It is indeed. However, while some people believe users are unlikely to share their spare capacity with people they don't know - the security implications alone are huge - others point to the illegal P2P network as a potential model. After all, music sharing started purely with some people wanting to share their record collections with others with no view to making profit.
Comments
There are 7 comments. Join the discussion
1. anonymous
What about www.grid.org - this is possibly the largest group of number crunching volunteers, 1,226,683 of us representing practically very country in the world, searching for a cure for cancer (sponsored by The University of Oxford and the NFCR). Members are also working alongside the WCGrid on the Human Proteome Folding Project.
2. David
I too would have started with the likes of www.grid.org
(If you have not signed up... then please do)
And I like many others have donated spare computer capacity to the 'Cancer' project over a number of years.
What I would question is how these grids will fair when the projects are not what I would term a good cause..... I for one do not have a problem with finding a cure for cancer or helping in some way to develope drugs that may help ease the lives of people suffering from illnesses but beyond that I would want cash reward for crunching for Companies and such.
In the end there really is no such thing as a free lunch, is there?
3. Dion
One glaring ommission in your otherwise somewhat informative notes.
How does a grid differ from a cluster?
4. anonymous
These guys are a little bit in the dark ages! - We've been working for some time with a prototype product called OmniFabric, (www.omnifabric.com)
This amazing tool moves grids into the application layer and delivers real benefit in terms of both reduced CPU and massively reduced lead times - I understand they're going for a launch shortly - you guys should take a look.
5. Rich Wellner
A Grid differs from a cluster in that it is a superset of cluster functionality. A Grid contains not just a lot of CPU, but also data transfer capabilities, storage and security which make is suitable for computing in the large.
6. Rad
Grid vs Cluster: Just to add more, further to Wellner, Grid is heterogenous and Cluster is homogenous, I suppose
7. Ant Evans
The phantom processing shortage
Grids work by harvesting a free abundant resource - idle processors. To get to those processors, they distribute work over a non-idle network. Because networks are more expensive than processors, the successful public grids are all processing-heavy, throughput-light applications. They are suitable only for jobs with tiny data requirements.
The economy of a local network is different. The ratio of spare processors to spare throughput is orders of magnitude lower. Fine: most business applications are throughput-heavy. Unfortunately, this increases processing and storage overheads, because more data means more parcelling and caching. So commercial intragrids' success will rely on finding applications that match enough spare cycles with enough spare pipes with low enough overheads – all to avoid buying more of the cheapest resource on the network.
Public grids are already lowering the cost of doing science, but that is because of their non-businessy economics. Private grids are likely, likewise, to stay specialized.