Invest in diesel generators. Why? According to Paul Kedrosky (venture capitalist) and Rich Miller (who keeps a blog about data centers!) the big constraints on building data centers isn’t any of the obvious things like availibility of contractors or whatever but that there is an 18-month lead time at the moment on 2megawatt diesel generators - and you have to have them as power backups. Or buy land where energy is plentiful and cheap, because that’s where all these big companies are going to need to build their data centers. Ten years from now, the big names in energy will be companies you’ve never heard of now. GE will survive, though, because it has a long history of survival.

Economics encourage people to waste things that are cheap. Currently, what’s cheap is processors. So people put in thousanjds of things, only to discover that the power to run them isn’t cheap - over the last 3-4 years power consumption per rack has gone up 3-4x. Plus the numbers keep going up - Google expects to have 800,000 blades by 2011. And even if you start installing more energy-efficient stuff now (Intel and AMD are both making pitches that their dual-core chips are more power-efficient than the older generation) it will be years before all the legacy stuff is replaced. (He likened this to old cars on the highway; but in fact the cars of the late 1970s and early 1980s were often more fuel-efficient than today’s fashionable SUVs. I have lots of friends that get worse gas mileage in their 5yo cars than I do in my 20yo Nissan Prairie.) It strikes me as ironic that just as everyone in the UK is getting all fired up about being green and energy efficiency they are all at the same time installing air conditioning and leaving more and more computers switched on all the time. One point Alec Proudfoot (of Google) made is that a lot of savings can be achieved by improving energy efficiency - eg, an ordinary PC PSU is about 70 percent efficient. Increase that to 85 percent, and you can achieve megawatt savings without turning anything off.

Which reminds me that yesterday over breakfast UK journalist Jon Honeyball told me that there is an incredible vulnerability in London relating to the path the pipes take to the main area where everyone’s disaster recovery backups are stored. Kind of like the fact that the main pipe carrying water to LA goes over the San Andreas Fault.

All of which goes to show that the systems we’re building are vulnerable to all sorts of non-obvious constraints. Like the piece Simson Garfinkel wrote in Wired some years back about “Fifty Ways to Kill the Net”. Number one was: “Buy ten backhoes.”

wg