Two issues that burned me a couple of months ago with EC2 and Elastic Block:
I built my server and got it running just right. I then wanted to back it up, so I bundled it (save your instance to S3) but forgot to register it(list your S3 bundle as a privately available AMI). When I terminated my instance I wasn't able to restart it, because it wasn't available in my list of registered AMI's. I had to rebuild my server from scratch.
The second problem that I had was in rebuilding my database after my instance terminated. Postgres keeps the log files (transactions the database has completed) in the install folder and the data file(what the database has stored) in the storage folder. When I terminated my instance, I lost all the log files that were stored on the EC2 'disk'. I had the data file on elastic block, but the log files were gone. The standard recovery process for postgres needs log files.
It took me a day to figure out how to rebuild the database from just the data files. I now have both the database and the log files on my elastic block image. Hopefully that will help the next time I kill my server. :) I also now have automated backups of everything.
How are you running your automated backups? Have you automated deployment too?
As far as EBS, another thing to think about is instance types. Since EBS is essentially network attached storage (I believe), network IO greatly affects performance. Small instances don't have particularly great io. Thus, I prefer to run off the transient disk right now over EBS. If we went to larger instance types, which have better network io, that might swing the pendulum.
I have always wondered about how EBS as disk will affect database as it is essentially a network drive. Any info on that would be great.
The Solaris performance was really poor compared to Linux. I emailed Sun and initially they were really receptive but never explained what was happening. I'm hoping to re-run the tests again soon to see if the latest OpenSolaris (2008-11) has improved.
You can run MySQL off of the transient disk ("/mnt") but you should consider running it off of an EBS volume. One reason is because you won't lose the disk if something happens to the instance. Another is that you can use the snapshots feature to create instant backups of your database. Later on you can create new EBS volumes from a snapshot (which is really cool for building a stage environment with production data).
This may sound a bit complicated when you're just starting out. It was for me anyway so after working with AWS for a couple years I started working on a development and deployment tool for EC2. The configuration is entirely via a Ruby DSL.
You define the properties of your environments and roles with the machines DSL (including EBS volumes) and you script your repeatable processes (like deployments) with the routines DSL:
http://github.com/solutious/rudy/blob/0.7/Rudyfile
If Python is more your style, there's boto: http://code.google.com/p/boto/