There's a lot of "probablies" in this write up. I applaud the reverse engineering attempt, but this is a little hand wavey.
I totally agree. The intention of writing this post was not to provide a precise documentation of Amazon's architecture (only Amazon can do this), but to get feedback if I am on the right track to understand the underlying architecture.
Thanks for the work on this and for writing it up. Did you try engaging Amazon to get some more direct answers? Are they not forthcoming about the infrastructure?
It would be really cool to be able to get in-depth about the hardware setup, how the disks are clustered and interfaced with the CPU-heavy clusters, etc.
I contacted someone at AWS through a friend and asked him basic architectural questions, like if he can confirm that they are running Linux in Dom0. However, AWS can not give such details as long as you do not have a valid business need and sign a NDA, which I declined. If they are so closed about basic questions about their architecture, I doubt they will give in-depth details.
It'd be nice to see something like this done for linode and rackspace as well.
Agreed, but I think there's not as much need with linode & rackspace. I believe they pretty much work as you would expect, and they're sufficiently open that if you asked questions my guess is that their default position would be to tell you (in contrast to Amazon.)