Rupert Goodwins' Diary

Thursday 24/11/2005

SETI@Home was fun, but we didn't catch any aliens. The project is now being moved from its custom client to BOINC, a new distributed computing architecture that lets you pick a variety of projects and assign priorities to them. Meanwhile, IBM and others are getting together to do AIDS research on its World Community grid, run on much the same lines.

All this is fine, but many people quail at the idea of running other people's software on their computers. It uses power and introduces security problems. Ideally, you'd want the grid software to run in its own little virtual world where it had no access to anything other than CPU — or if it did, then it was heavily managed — and where its power use could be metered. That way, a user — or even a corporation — could build up credits which could be spent buying CPU time on the grid if they needed it, or for other services.

It could also establish a market in energy. IT takes up an increasing proportion of industrial and consumer electrical consumption, which means it can be used for load balancing. Every electrical generation and distribution system suffers from daily variation, where it has to be capable of supplying peak demand but efficient at lower levels. It's much easier to build an efficient system that has a constant load — that's the idea behind Economy 7 and the like, where people are encouraged to use power during low-demand periods to even out the peaks.

With energy-conscious computing connected to a grid, a server could be aware on a minute by minute basis where the most attractive place on the planet for processing was, and could cut back on its own power requirements during local peaks by sending off its processing to be done somewhere where demand was otherwise low.

It could be a lot more important than that. With the latest data from environmental scientists saying that we may be much closer to the point of no return than we thought, where various feedback systems take over and the world changes at catastrophic speed. At this point, we stop shaping the climate and it starts to shape us. The global economy will have to change dramatically, and how easy or hard this is for us will depend on how well we can model what's happening to the point we can predict what's up next.

I can think of no way of modelling environmental change accurately that doesn't involve a great deal of distributed processing linked to local sensors and other systems — nor an efficient way of managing an energy-poor economy than lots of local control over who's using what. Grid computing will be absolutely key to that, and our initial experiments show we need to take it very seriously.