Cambridge brain touts wind-n-server combo farms
Thin-client spinoff bizprof spins skinny as green
The head of Cambridge University's Computer Lab has advocated a restructuring of today's computer architecture, in which processing power would move away from desktop systems and scattered data centres and be centralised close to mighty wind farms.
Professor Andy Hopper's green thin-client plans were reported in today's Guardian. The prof outlined his ideas in a speech to the Royal Society yesterday.
"I think it is very interesting to contemplate a world with a smallish number of server farms, huge ones, which are deployed in places where the energy is produced," said Hopper. He said it was much easier and more efficient to transmit information across distance than it was to transmit power.
The professor's scheme, according to the Graun, "would work by shifting computer operations to servers close to wind farms that are working at full tilt".
This would seem to involve having a lot of spare computing power standing about unused, waiting for the wind to blow in its physical neighbourhood. On the other hand, large amounts of the world's current processing power sit idle much of the time, particularly in desktop machines which are often switched off and overspec'd for the tasks they typically carry out.
Thus far, rather than having many data centres and switching between them to chase power, the trend has been to physically move the machinery to where the cheaper power is available. This would not, of course, be responsive enough to deal with wind fluctuations.
As for thin clients, that idea has been pushed before without massive success - notably by Professor Hopper, in fact. The good prof is actually chairman and co-founder of Adventiq, which sells KVM-over-IP kit, and of remote control firm RealVNC. He is also co-founder and director of gigabit-net firm Solarflare. All these companies could be expected to benefit hugely from a big shift to centralised hardware and skinnier user machines.
There's obviously nothing wrong with betting on your own ideas, but Professor Hopper is not exactly an unbiased commentator here. And there are many greens who would question the ecological soundness of his ideas, arguing that, in fact, the way to economise on power-transmission losses is to distribute the means of generation - which would tend to call for distributed processing. (Always assuming you regard IT power consumption as the pressing issue, which doesn't seem all that sensible to many people in the context of transport, heating, manufacturing etc).
Then, of course, nuclear advocates would suggest that - if transmission losses are really your main worry - you could simply have one server farm next to a nuclear plant rather than several at different windmill clusters in order to deal with weather fluctuations.
And so the merry eco-debate day winds on. ®
Sponsored: Hyper-scale data management