I'm a big fan of using EC2 spot instances to help reduce costs, but the pricing behavior can make managing them a pain. There's a very bimodal distribution, where the price will wobble along at around 16 cents an hour almost all the time, with infrequent but sudden spikes up to 50 cents or so, above the non-auction cost. I don't know for sure why things are so erratic, but I can think of a couple of possible causes.
It could be that there's a very non-linear distribution of maximum bids, where almost everybody is willing to pay up to the normal price for an instance, and so when capacity is reached and machines need to be shut down, the price has to shoot up before any significant number of resources are freed up.
There could also be some very heavy hitters who occasionally demand very large numbers of machines, so causing the price to spike. What was interesting this evening was that I was still able to manually start up a couple of instances at the regular price, which was below the auction price. I also find it surprising that the two data centers seem to be so highly correlated. Unless there's sophisticated users who are quickly switching their requests between availability zones, I'd expect a lot more independence as the spare capacity in each varies.
Of course, there could be something deeper at work. Werner Vogels was born in Europe, and the Swiss National Bank is trying to find somewhere to sink all its money to prevent the currency from appreciating. Coincidence? Mind you, I talk a bit funny too, so it's hard to trust anything I say. Unless I'm narrating a documentary about meerkats, in which case the accent helps.
I had always wanted to learn about this topic … I think it’s great the way you expose .. great work and continuing on with this great blog
Major thanks for the blog. Really Great.