It's about damned time. They've been trying to pretend EC2 is for utility computing...like running large distributed apps for a short period of time. But nobody is using it for that...people want Amazon's network and a cheap way to scale from zero to huge for websites. Now that they've finally embraced that, if they'll just resolve the reliability issues, EC2 will be a fantastic hosting option.
Lots of people actually do use EC2 for utility computing, often powered by Hadoop. Most of these folks aren't especially vocal about what they do, but there's data mining on a large scale, drug discovery, and similar stuff happening inside.
Heh, thinking of exactly that, I was going to write about how turkeys are fed and well taken care of by humans for 99% of their lives, until that last day, when things take a turn for the worse.
That said, things will run just fine, even with downtime, if you're not depending on the machine in any way, don't have any local storage, and are capable of switching over to other systems. However the lack of downtime might lull people into not worrying about that stuff.
You raise an important concern, but I wonder if the analogy is the right one. For a lot of startups/businesses, a period of downtime (in hours, not days) is not death. It might suck, but you won't die. In the late '90s, Ebay was down for three days straight (!) and it's still doing OK.
I will be surprised if Amazon doesn't work out the reliability issues anyway, given the massive resources and smart people it has available to it.
A bit of downtime is certainly not a catastrophe, but it might turn into one if you lose a bunch of data because you weren't properly shipping it elsewhere.
A much more sane response to the overwrought fears of EC2 downtime. If you don't know your hourly/daily income rate (often $0), then you can't even correctly value uptime.
Thats just it. Downtime is expensive if you rely on your web site to make you money. If you are a small eCommerce shop with all your eggs in the EC2 basket, downtime can be expensive.
I still think they need to resolve the disappearing disk issue, or continue to work on database type things like SimpleDB for persistent storage. Something using an actual relational DB like Postgres would be even better.
For me, the biggest step in resolving the reliability issues would be for Amazon to provide load balancing in the API. With a static IP, I could run my own LVS-based load balancer, but I might screw it up. If Amazon provided load balancing, their other customers would help shake out the major bugs, and I'd expect it to work much more reliably.
