The data persistence problem is annoying. For databases you can setup a Master and slave, the chances of both going down are slim so if one goes down you bring up another and replicate the missing data to it. You can also do a full nightly dump for extra insurance. (At this point in the beta stage I would probably recommend it from the instances ‘disappearing’) but, I have no first hand experience.

As for DNS I’ve seen people post about using round robin dns load balancing. Writing scripts that will update their DNS provider with “add me” to the rotation and pinging each other, then when one goes down it removes that server from the rotation. Poor man’s load balancing, but its a semi-decent solution. The only show stopper is if you need SSL. Obviously SSL needs to bind to an IP and if an instance goes down and you can’t get that IP back, you’re screwed!

As for keeping the AMI Generic, I would install ALL the services I need in an AMI and based on ’startup’ metadata it acts as a specific service. So you say ’startup.sh master-database’ the instance starts up, grabs the master-database configuration you have living on S3, it modifies the instance to act as a master-database. — Yea I know this takes a bit of setup, but it would be an awesome setup when its all done. =)

So, for me, the current reason for not going with EC2 is static IP for SSL. I’m sure they are working on a solution, but it stinks to see this awesome system, but can’t use.