Friday, February 17, 2006

Intellect "After 5"

Last night I attended an Intellect "After 5" briefing on Grid Computing, co-sponsored by the Welsh Development Agency. I did learn a few things. For instance, who knew the Welsh made wine (and that it was surprisingly not bad)? Or whisky (not tried yet)?

The actual stuff about computing was very general. The first half consisted of an overview of "Grid computing" by a couple of nice blokes from ThoughtWorks. Ben and Sam focused on computational grids (like SETI@home) rather than data grids. Hence no mention of Google or Amazon, of Sun's utility computing model or even RAC. Still it was interesting to learn that Novartis are using the scavenging technique to assess candidate drugs. By downloading a screen saver program to 2700 of their employees' desktops they built a computational grid equivalent in crunch to the fifteenth most powerful supercomputer in the world. Apparently there are licensing implications for such an approach (or maybe they were talking about RAC after all!)

The second half was some evangelising on "globally distributed endeavour" from Dr Alex Hardisty of Cardiff University. Again it was interesting to hear about things like the medical community's response to the SARS outbreak in Taiwan but I don't think Alex quite made the link between experiments in collaborative working and Grid computing in a commercial environment. This felt more like Video-Conferencing 2.0 than anything else. I don't doubt that he is correct in saying that the future of the internet is going to be more about collaboration than anything else. But I think he underestimates the scale of the task. The real problems are not technical they are political. When was the last time you heard the phrase "joined-up government" being used in any way other than ironically?

I suppose I'm not the target audience for these Intellect briefings. Intellect's remit is to raise general awareness of IT matters in industry generally so the briefings are very non-technical. Nevertheless, I feel they missed a trick by failing to cover data grids adequately. Computational grids fit relatively narrow problem spaces: lots of repetitive calculation over discrete chunks of data. Most commercial users are likely to find data grids more applicable to their present needs. That's why it would have been useful to have more discussion about the IT behind Google or Amazon, both businesses where the scalability and performance of grid computing are key to their successes.

The other thing it would have been useful to learn about is progress towards the commoditisation of grid computing. The few examples of successful grid implementations I've heard of tend to be domain specific projects - hand-rolled, hardwired, hard coded. It's really only going to take off when the solutions become truly generic. Does that sound like RAC? Hmmm....