Commoditizing the grid

Grid computing has made a slow, but steady, entry into the consciousness of the typical business user over the last four or five years. Often seen as a niche technology, usable in only a few very specific markets, Grid computing is now seen as another alternative to the problem of maximising computing power.

In a nutshell, grid computing distributes work over a number of machines. For the niche markets, grids are used for everything from computational grids for proteins and fluid dynamics, right up to the use by studios like Dreamworks which use grids to render the animations and special effects in many movies.

Today, we're seeing grids being used in other areas, such as reporting (spreading the load of accumulating data into a single report) and even, through web services for more mundane jobs but in a flexible environment.

Returning to the movie studio use, the original Toy Story film - the first feature-length CGI animation - was rendered on a server farm using Sun SPARCstations. At the time, I had one of those machines (well, the same model) on my desk and it certainly impressed my fellow workers that the machine was so capable. The key, however, was not the power of the individual machine, which today would look pathetic, but the use of many machines in a grid-like environment.

This week, Sun made their grid-computing facility to all. You can run your application on AMD x86 or SPARC CPUs and be charged just $1 per hour, per CPU. That actually sounds comparatively inexpensive, especially when compared to building your own grid, and you don't have to worry about electricity or management costs either.

Is it possible to commoditize grid technology in this way?

Why not?

We are moving towards a services environment - Google, Amazon, Yahoo and others have already shown that people want a service platform, which they can either use directly or that they can adapt and combine with their own applications and technology.

There's no reason why we can't do that with raw computing cycles.