Party like it's 1999
Is what the AWS cloud is like. Too much cookie cutter stuff on the IaaS side.. PaaS is more flexible.
I remember meeting Andy and his "chief scientist" (forgot his name) a few years ago at their HQ where we politely vented our complaints about their shitty cloud. And got back the typical answers "yes we know that..and that..that will be fixed in the near future/next version" over and over.. the folks I was with said it wasn't their first visit(though I think it was the last) - and they had gotten the same excuses before. The CTO of my company after the meeting about shit himself saying he'll never take me back there again - I told him I was holding back, I really was, I was fighting the urge to just go full porno on the whole topic but I did not. My boss who was there too agreed he said he could see me constraining myself from unleashing the real honesty.
To this day they have not addressed pretty much any of the concerns we raised time and time again more than three years ago now.
They just operate so differently, they don't know/care how most of the rest of the world operates - if your not doing things the same way they are then in their eyes your doing it wrong. It's sad they are so arrogant about such things. Built to fail puts an inordinate amount of pressure on the customer that need not be there with the technology that we have today.
I tell people now using EC2 for anything other than this one use case:
* Highly dynamic environment (constantly scaling up and down on a daily or even hourly basis)
* Highly resilient application that can work in a built to fail environment.
Then your using EC2 for the wrong stuff. You may still be able to get it to work -- (the counter example I give is using HP 3PAR arrays for Hadoop storage), but it likely won't be pretty, cheap, or easy to use. At least using 3PAR storage for Hadoop while it won't be cheap - at least it'll be pretty, and easy to use. (I know a customer that uses HDS/BlueArc for Hadoop storage after having a terrible experiences with hardware failures on whitebox stuff..so they went the other extreme with enterprise storage - so there are folks out there that do it!). Hell you could go full blades + vmware +3PAR storage with hadoop. It'll be costly - but it'll be easy to manage(as in fully redundant), and it'll perform pretty well too (probably not to bare metal performance but it won't be bad either). Not saying it's the right solution for something like Hadoop - but it's easier to make that work, than it is to make EC2 "work" for just about anything else.
The last two companies I have worked at both deployed their apps from day 1 in EC2 cloud (current company moved out 18 months ago) and in neither case did the application support built to fail, nor was the scale constantly changing(end result = drive me to the brink of insanity). Things were static for months basically.
Companies before that even before using EC2 situation was the same - no built to fail, no constant changing scale requirements. Static load for months (except for growth etc). Same goes back over a half dozen companies and a decade+ of experiences both personal and from friends. . situation is very common.
The use case that is best for EC2 is the .001% use case.
I talk to companies on a regular basis that either have moved out (of EC2 or other clouds though obviously EC2 is the dominant player), are planning on moving out, or want to move out and don't know how. Unfortunately stories like this rarely make the news...it's always people moving "into cloud" not the other way around.. really REALLY frustrating. The reasons why are somewhat obvious - high costs, poor performance, poor quality, bad support. They(management folk) don't realize what they are getting into when they sign up.