Not a fan of RDS. It has all the same drawbacks of using an EBS partition for storage, but as far as I know you have no visibility (like await, iowat etc) to know when your storage subsystem starts to flake out. EBS removes a lot of Sysadmin & DBA headaches, but when you suddenly get mysterious DB stalls, good luck trying to figure out what's wrong.
I've worked with some systems that used a software RAID10 array of EBS volumes, and when a single EBS volume started to flake out, they'd just kick it out of the array, and performance would improve. They'd then rebuild the failed replica during off hours. If you're going to rely on EBS for DB storage for anything in production, this seems like this is the way to go.
I second this. While EC2 takes a bunch of really hard problems, allowing you to provision virtual servers at a moment's notice, the value proposition for RDS is much more marginal. It's simply not that hard to get MySQL up and running, even in a RAIDed configuration, and the fact that the underlying store is EBS means most of the advantage of RAID is spent simply overcoming the I/O handicaps of EBS.
We prefer running our MySQL instances on ephemeral storage, which is physically closer to the instance, and deal with I/O volumes by using lots of replication.
The article about the new MacBook Pro line also included a discussion of Airs and Mac Pros. Sometimes the discussion strays beyond the boundaries of the headline.
Most of the time that is a good thing - I for one am interested in the point the OP raised, which is that RDS (at all instance sizes) gives you no visiblity in to what is happening under the covers on your DB server.
He's informing us of the drawbacks of using the other plans. This special plan would be (for some) dipping toes in the water of RDS only to find out its infested with piranhas.
I've worked with some systems that used a software RAID10 array of EBS volumes, and when a single EBS volume started to flake out, they'd just kick it out of the array, and performance would improve. They'd then rebuild the failed replica during off hours. If you're going to rely on EBS for DB storage for anything in production, this seems like this is the way to go.