"A Grid is a collection of distributed computing resources available over a local or wide area network that appear to an end user or application as one large virtual computing system." - IBM
"Conceptually, a grid is quite simple. It is a collection of computing resources that perform tasks. In its simplest form, a grid appears to users as a large system that provides a single point of access to powerful distributed resources." - Sun
"[The Grid] lets people share computing power, databases, and other on-line tools securely across corporate, institutional, and geographic boundaries without sacrificing local autonomy." - Globus Alliance
"Grid computing is computing as a utility - you do not care where data resides, or what computer processes your requests. Analogous to the way utilities work, clients request information or computation and have it delivered - as much as they want, and whenever they want." - Oracle
Grid computing has been so heavily hyped up in the technology press this year that most major players in the software arena have staked a claim in what they see as the 'next big thing'. As you can see from the definitions above, no-one seems to be quite sure what they are working towards. Indeed it could be said that there seems to be more activity in arguing over a definition than there is in developing the technology.
So is the Grid just marketing hype or can we expect to see some real benefits from these ideas?
To tackle this question, first of all it's necessary to look at the types of application that businesses will be running on the computing platforms of the future.
Of particular interest is the ratio between network usage, processing time and disk storage required for a particular task. The following example explains why this is the case:
Currently, one pound will buy you 1 GB of internet traffic, 8 hours of CPU time or 10 million database accesses.1
The SETI@Home project has so far used 1,643,925 years of CPU time, donated by millions of computers around the world, searching for patterns or signals in radio telescope data. Using the above figures, to do this in a traditional manner would have cost about fourteen billion pounds.
However, due to the nature of the task at hand, SETI parcelled the work into about a billion packages and sent them out to people volunteering their spare CPU cycles. As each task could be described in just 0.5MB of data this required a total network bandwidth of 500,000GB which cost them about a million pounds.
A fourteen billion pound calculation for one million is a pretty good saving!
However, this relies on one key factor about the SETI calculation: the work could be sent out in 0.5MB parcels and each parcel would represent about 14 hours of work. It is this ratio of CPU cost to network cost of 10,000:1 that made SETI@Home viable. It is worth noting that this is not a common feature of many tasks that businesses will want to perform.
Most business-related calculations that could benefit from the huge computing resources that the Grid promises rely heavily on access to large amounts of proprietary information. The fact is that the cost of shipping this information across the network will immediately negate the benefits of having someone else manage the processing resources for you.
For example: a marketplace simulation requires, by its very nature, a high degree of interconnection between the entities that are carrying out the different parts of the calculation. Many, (if not most), real-world tasks simply cannot be broken down as easily as the SETI calculation.
So how does Oracle, a database company, justify calling its latest database 10g (where the g stands for Grid)? Why does Sun believe that its business customers will benefit from Grid technologies?
In the case of Oracle, they are just referring to a technology that has been around for a while called clustering. Clustering simply connects several computers together to act as one. This is nothing new and Grid in this context is nothing more than hype - Oracle are just using a topical 'buzzword' to market an already-existing technology.
Sun, on the other hand, may have a more reasonable position. It touts intra-company Grids rather than ones based on a worldwide reach. In these cases, the above analysis breaks down - as the cost of network transmission inside a company is a tiny fraction of that over the Internet. However, there are still only a small number of applications that could benefit from such a configuration.
That said; there are a few notable exceptions. Smaller architecture firms, which lack the huge computing resources required to ensure the safety of more ambitious projects, will be able to have their designs tested using Grid computing power: the input (a technical drawing) and output (a structural analysis) are tiny compared to the massive amount of processing that needs to be done.
Similarly, 3D animation studios could render movies using the Grid; as the 3D model input and the picture that is the output are again small compared to the processing required.
This article is not to say that computing Grids won't have an impact on our lives - high end scientific projects such as cancer research, human genome research and development of new materials will benefit hugely from the huge pool of computing resource that the Grid promises.
In the commercial world however, the Grid will only impact the very largest of enterprises until network connectivity is as cheap and readily available as processing power. With Moore's Law2 still holding, that is not likely to happen any time soon.
To summarise then, the computing nirvana of processing and storage on demand (also known as utility computing) is a long way off - both technically and, most importantly, economically.
If you would like any more information, please contact FWOSS - we'll be happy to discuss our views with you.
1 Distributed Computing Economics Jim Gray March 2003
2 "Processor speed will double every 18 months."