Rupert Goodwins' Diary

Friday 16/6/2006

To the BBC World Service, where instead of speaking peace unto nation I'm asked about Google's enormous yet secretive new data centre being built in a remote part of Oregon. Yes, I know that's a tautology. To which I can answer anything I like - Google's not saying. It's put the whole nearby town under NDA, and there's no point in spilling the beans while Microsoft and Yahoo are busy spending their money trying to build data centres to catch up

Some things are obvious: it's out in that part of the world because of plentiful, cheap hydroelectric power. If it's going to grow enough to satisfy its investors, it will need acres and acres of servers. Some things are pure guesswork: thin client, real time language translation, even the fun Google's had with us and its hints of artificial intelligence and biological data sifting.

And more than that, one cannot say. Will it show up on Google Earth? Is it a good idea to build everything into one huge centre, where it's vulnerable to physical disaster or some nutter in a nicked plane? Is it indeed all eggs in one basket: how many other clutches are in smaller baskets distributed around the planet?

A distributed system makes a lot of sense, for many reasons. The best example is the Internet, which despite being the most complicated thing we've ever spun, despite being in a constant state of reinvention and spurts of exponential growth, despite being used by millions of people for all manner of independently generated and unsupervised purposes, hasn't stopped working in twenty years.

It also helps ease the big, big problem of data integrity. If you build a data centre the size of Wembley Stadium, how do you back it up? It'd be bad enough if Google Search dropped off the edge of the world - how much worse would it be if we're all relying on Google thin client to look after all our data and perform all our computing? A big fat data centre needs a big fat pipe or two, which is vulnerable to mishap or mischief: a distributed data centre has thousands of connections, and isn't.

Google has been running hundreds of servers around the world for years, so it knows all this. In fact, I suspect that the data centre is probably not so much the Fort Knox of the modern world as the red herring of the punditsphere. But I wouldn't mind a look around.