This might be what you mean, but a combination of memcached and a replicated database running on some ec2 boxes should work pretty well. The socialtext link implies that.

I haven't used simpledb yet, but as far as I understand it seems like it could be a good use case for replacing the replicated database piece.

All you need for your backing store is a dumb key/value dictionary. You want S3 for that. Even with replication, running a DBMS on EC2 is a dreadfully foolish idea. A lot of startups are going to immediately go under the moment that EC2 has its first system-wide crash, and I'm going to be laughing at their misery.

Here is the thing. What makes some other solution more immune to a system-wide crash than ec2? When that happens, what is the major difference between that solution and an ec2 based one? If you really want to play it safe and don't trust amazon, then what you want is a disaster recover site which is completely independent. However, wouldn't you also want the same for any isp/hosting solution?

It isn't a question of who you trust. It's a question of what classes of events that your system can withstand. EC2 is not durable. It isn't designed to be.

That's considered to be a bad idea generally.

http://www.socialtext.net/memcached/index.cgi?sessions

But memcached + s3 as the backing store could work. Or for a low traffic site you might just be able to use s3 directly.