Alex Steer

1145 words | ~6 min

I'm still amazed how much of the public conversation about cloud computing has been focused on the upside. Admittedly, the upside is pretty good: the promise of reliable, scalable, cheap computing services, where somebody else does the grunt work of maintaining the servers and connections, and you just pay for what you use, like a driver on a toll road.

But it's good to see some of the downside risks start to make their way to mainstream attention in the wake of the Amazon EC2 service meltdown and the Sony Playstation Network hack. Bloomberg does a nice job of drawing the line between those two, showing that EC2 was used to bring down the Playstation Network. Is that the first case of cloud-on-cloud crime?

Reliable, scalable, cheap computing services, people are starting to notice, may suffer from a Holy Roman Empire problem: they may be neither reliable, nor scalable, nor cheap.

It's always slightly unfair to bash the reliability of a whole infrastructure because of some significant failures. Neither Amazon nor Sony means cloud computing as a model is necessarily flawed - but we should avoid the temptation to use the 'one bad apple' excuse. In these cases both Amazon and Sony were unreliable. That doesn't make all cloud computing unreliable, but it may show that organizations that rely on it suffer from low resilience.

Here's the difference between reliability and resilience. If Bob's Widget Company outsources all its computing needs to cloud providers, it probably gains computing reliability. As Adam Smith might tell you (and it's not often I cite Adam Smith), professionalization tends to drive up quality at least somewhat. Bob's decision to let computer professionals handle his computing needs is smart. In the hands of a pro, his services are less likely to topple over, and more likely to be fixed quickly if they do.

Economy-of-scale logic kicks in here. Rather than hire his own systems administrator, Bob can outsource to a big shared hosting company, who can do this job more cheaply with little apparent drop in reliability. For Bob, this is clearly an efficiency gain. He may also think of it as a gain in resilience. The economy-of-scale gains and the competitiveness of the shared hosting industry means lots of good things like better backup systems, version control, proper firewalling, etc. Bob's computing services are in safe hands.

So where's the drop in resilience? Well, it's to the resilience of the whole system. A thousand computing services hosted in a thousand locations is less efficient and reliable but more resilient. If Bob's and his widget-making rival, Tim, host their computing services in different places, and Tim's services are hit by a power outage or a bus crashing into the office, Bob isn't harmed, and may gain from the extra business from Tim's customers. If they're both in the same place, along with the services of lots of other widget-making rivals, then it's goodbye widgets.

In a low-resilience system the ripple effect from failure is greater. We've seen that with Amazon EC2, we've seen it with the attacks on Tumblr and Paypal, and we've seen it writ large with the subprime mortgage collapse. In that last case, in particular, it wasn't just the concentration of a single risk that caused so much trouble - it was the interaction of lots of risks. The assumption that all subprime mortgages were, like the unhappy families in Anna Karenina, each unhappy in their own way, led to an belief that a risk to one would not be a risk to all. Which meant that, a little like cloud services, all those mortgages could be bundled together into derivative products whose risk of going bad would supposedly be less than the risk of each individual mortgage defaulting. When the waves of foreclosures happened, it became clear not only that the risks to thousands of subprime mortgages were all much the same, but also that the interaction of risks - between mortgages, CDOs, credit-default swaps and all the other glamorous products hosted in the mortgage cloud - created a downward spiral effect, hastening the collapse.

Even more than shared hosting systems, cloud computing - where the underlying services are more fundamentally shared - are low in systemic resilience. And this means that all the talk of their scalability may also be pretty risky. They can be scaled in much the way that I can stack pennies on top of each other without much effort. I can keep on and on until I have an impressive tower of pennies for very little incremental cost. But if they fall over, I've got a lot of clearing up to do.

What about 'cheap', then? Again, this becomes a conversation about complex systems. The incremental cost of cloud computing to the end-user is extremely small. But that's because we're operating in an economy which assumes that a lot of the systems that let you access cloud services will continue to be extremely cheap, incrementally almost costless. As my colleague Andrew Curry has pointed out:

The technology industry has grown up in an age of cheap and abundant energy, and that has shaped, deeply and fundamentally, the way it sees the world, what it chooses to make, and how it designs what it does... But the age of cheap and abundant energy is coming to a close. It is about to become scarcer and more expensive.

The cloud computing model assumes cheap electricity and cheap bandwidth, assumptions that pretty much no scenario I've seen supports. This poses two problems: one about business continuity, the other about marketing. If the energy outlook or the bandwidth cost becomes more volatile, mass-scale cloud computing may suddenly look less attractive, since despite its low incremental cost it adds additional layers of cost in terms of data and the need for persistent high-bandwidth network connections.

The second problem is about changing attitudes. The available energy forecasts depend for their success on a reduction in resource use, especially in developed markets. After the economic downturn reduced disposable income, people learned how to tighten their belts, became more savvy and calculating about spending, and have grown angry with businesses and brands that seem to promote wasteful spending and financial irresponsibility. Might an energy downturn not prompt a similar anger at business and brands that seem to promote an over-indulgent approach to data and downloading?

So my question for marketers is: could we be heading for a cloud crunch? And if we are, could your enthusiasm for the cloud and for the always-on world turn from a strength into a vulnerability?