Server platforms Toolkit

Get your hands off our grids

Leader ZDNet.co.uk

Published: 20 May 2005 12:30 BST

When the man at the centre of grid computing development in the UK sounds an alarm, the industry needs to sit up and take notice.

Grid computing is a peculiar beast — unlike most big enterprise technologies of the past few decades it hasn't been driven by the vendors. Instead, it grew out of academia. It's also different because it's not supposed to be about selling more kit or even more software — much of the code you need is freely available.

Rather, grid should mean making better use of what you have; clawing back those spare processor cycles and dark hours of downtime. Universities, with their vast networks of PCs and servers, have both in abundance.

So what's the problem exactly? Well, grid's gone and grown up and got sexy, and now every marketing man wants to gets his grubby mitts on it.

As a result, what is often actually clustering is being marketed as grid computing, just because it sounds better. Sometimes it is called autonomic computing, sometimes adaptable computing, on-demand computing, utility computing or agile IT. So what's in a name? Quite a lot, really, when a large vendor is trying to pass off one technology as another.

Grids and clusters, for instance, do jobs that are equally worthy but also totally different. If you need to find every spare processor cycle to work on a large problem then grid computing is for you. But if you need load balancing and resilience then you're talking clusters.

Clusters work well on massively parallel jobs — think Google splitting search jobs up across thousands of low-powered servers — and where there is a balance between computing and I/O demand. The machines in a cluster tend to be topologically and physically very close. Clusters need stability, predictability and uniformity, none of which applies to grid. Grid computing works well for Sun shops where they have lots of workstations sat round not being fully utilised. Add some grid software and you have a compute farm, but a variable and unpredictable one.

Vendors need to stop trying to pass one tech off as another, just because it's sexy. Otherwise, they risk harming both grid computing and their own wares.