Amazon's ELB looks interesting, but I have a hard time evaluating how reliable it might be. I don't have a good sense of how it's implemented or what it might do behind the scenes.
If an ELB machine goes down that is proxying a request to one of my EC2 isntances, I expect that request to die. What about the next request that client makes? It had resolved an IP that used to pointed to some ELB endpoint that was on a machine that crashed. Is there a good chance the next few requests fail too? At this point are we at the mercy of a DNS TTL expiration and update?