[mmt]

> a different kind of cluster, that could optimize for better throughput, and more efficient storage

It doesn't even require all that much imagination, considering how anemic the storage IO choices are under AWS. Until recently, EBS was connect by 10Gb/s ethernet. Now, I believe the best is 40Gb/s ethernet, but even that's not ubiquitous and is only equivalent to single 4-lane SAS3 connection. Local (to the node) storage options are extremely limited.

The typical cloud provider model seems geared toward scaling "out", even before scaling "up" (while price:performance effective), which, of course, is a great way to maximize revenue. For a user/customer, it's terrible for cost and for performance, but this doesn't seem to bother VC-funded startups, for example.

Whether Oracle has the knowledge already or is able to hire that knowledge to create such a different offering is an open question, as is if there are enough customers who care enough to pay for it.