The economic implications of power

I was in a meeting with Cisco yesterday where they talked of the work they are doing with respect to greening the network in the data centre. Basically, they said, you can increase the efficiency of a data centre by decreasing the number of ports that each server uses. With virtualisation on the rise, that is apparently a big concern. The newest switches apparently take 10 watts of power on each side of the cable to run. Add that up over thousands of ports and you can see the point.

Actually, this is an ever-so-common story. The availability of power to data centres is swiftly becoming the limiting factor in much of what we do these days. Remember those old glory days where we had to optimise our systems to use as little compute and storage as possible because there just wasn't much to go around?

We are now in the same boat again - but now there isn't enough power to go around.

Gazing into the future a little, what is the effect of not having enough power to go around? Yes, the price will go up. Getting more requires large scale and long term investments in infrastructure, so to all intents and purposes the supply is fixed.

Is it so hard to envisage a situation where large corporate users of power are expected to kick funds into the consumer market to stabilise prices due to their disproportionate use? Or get charged a special carbon tax to offset their power use? Or actually wind up having to build their own power generation? I think it isn't.

Now, if that or any scenario similar occurs, we are faced with a very dramatic increase in our IT expenses- without any increase in capability. Can you imagine the impact on our cost to income ratios, which have been falling steadily over the last decade? They'll suffer a swift reversal.

20 years ago, the biggest cost when you wanted to start an airline was buying the planes. What is it now? The fuel. The entire dynamic of the airline industry was changed by it.

How likely is it that the largest costs of a particular IT capability - over the life of the system - will be the power that drives the system? I think very. Such a development will have architectural and economic consequences for everyone that needs a lot of compute.

Banks? We're practically all computer.

In deregulated markets, it is already the case that you can buy power in advance. For a fixed price. Having a futures exchange for power is a natural consequence of scarcity, so I suppose those will become more common in the near future.

But power is an input in a long value chain that results in data centre compute. There are so many other things that have to be taken into account as well: the cost and availability of network links, the skills of people in the area, and the economic environment of the local region. These all result in various - and fluctuating - prices per MIP in the data centre.

Wouldn't it be natural, then, that we might see the emergence of MIPS markets? A futures exchange where one buys cycles in the cloud? If we take the airline analogy again, that is certainly what happens when carriers need to guarantee the price of their fuel for the future.

I think it won't be that long until we will need the ability to run our workloads on any cloud, anywhere. That will be because we'll get our MIPS for the best price, in advance, on a futures exchange. Of course there are technical and cultural issues with this, and I can already imagine the security folk meltdown when anyone dares to suggest putting customer data not just on someone else's data centre, but on a random facility dictated entirely by price.

But economics are a powerful motivator.

Comments

[Kris]

"Amazon Elastic Compute Cloud (Amazon EC2) is a web service that provides resizable compute capacity in the cloud" - Almost sounds like you wrote the ad.

http://www.amazon.com/gp/browse.html?node=201590011

[Thomas Barker]

Have a look at this http://download2.3tera.net/demo/applogic20demo.html !

I'm sure there are caveats, but it's impressive that it works in general.

[Sam J]

A MIPS market proposal is the subject of an article I'll be posting shortly, along with a suggestion for breaking how it might all work (that is, a better approximation of the energy grid).

In the mean time I've just posted what I think comes close to a 'consensus' definition of Cloud Computing (complete with a 2,000 word explanation) that I think you might find interesting:

"Cloud Computing is the realisation of Internet ('Cloud') based development and use of computer technology ('Computing') delivered by an ecosystem of providers."

Cheers,

Sam