when random events aren't random - by Peter Judge
"It's just another random event", a colleague said to me when I sent him a link to the story that a car crash had triggered a power outage on Amazon's EC2 cloud service. But was it really?
In this instance, a car hit an electricity pole, near an Amazon data center, and cut off its electricity supply. Most of the site went over to generator power, but one of the transfer switches didn't work properly, and some customers lost service for about an hour.
Data Center Knowledge links this to a story about three other Amzon EC2 power outages. but does not draw any conclusions about the value or reliability of Amazon's cloud services - and nor do I. These things happen.
But are they random?
My colleague remembered that Amazon's cloud had some problems last year before with a lightning strike - an event which amused the Register.
One thing is clear - I'm more likely to hear about a data center problem if it is associated with a "big brand" name and some sort of real-world event like a car crash or a lightning strike. If I hadn't seen the words "car crash" and "Amazon" I'd never have clicked through to Data Center Knowledge's story about the other three outages (none of which had an interesting real-world cause).
There is nothing random about a series of events which are strained through the powerful filter of a journalistic sensibility. I'm betting there are numerous incidents like this at less famous data centers, with less obvious causes.
Apart from the nagging questions which Data Center Knowledge's readers ask (in the comments below both stories). Why did the switch fail? Was there sufficient battery power? Was there a full UPS?
The same questions apply to any other cloud service, whether its failures reach the public or not. Any data center is built to a price, and users need to know whether the right technology is put in to guarantee reliability.
However, the green point comes in here. Remember that the cost is not just financial. It costs in environmental terms too.
I've been looking at the power supply for temporary set-ups like press offices at events lately. My cycle home from the office takes me past the Houses of Parliament and College Green where, last week, an international media pack set-up temporary studios - in glorified tents - to cover our election.
All those studios had power from temporary diesel generators or "gensets". And all of them had some sort of reliability factored in - consider the loss in revenue if you had no coverage of genuine breaking news. Some of these media camps actually have duplicate diesel power, running full time to be absolutely certain of keeping the news flowing.
A long-term facility like a data center obviously has the opportunity to put in place more sustainable power supplies and more efficient ways to hand over to them. But there is always a cost.
Even a flywheel UPS, a very efficient way to store energy for a handover to the generators, does draw power. So reliability costs, in money and environmental terms.
At some point, I think that lower reliability options may be sold as greener because they use less energy - especially as greener will also mean cheaper.