The Eternal Mainframe is an interesting article. It was a great read because Clive and I have been saying the same thing for over a year. Quite a few more people if you add the "reinventing the wheel" thing in there.
I recall Bruce predicting that computers become a utility and security a utility function. Turns out McCarthy predicted the same thing about computers.
The funny thing is I've spent the last week or two looking up mainframes and minicomputers. I've looked at their history, the current mainframe makers, the hardware, the OS's, the software ecosystem. I think the current mainframes are often way better at the utilization, reliability and consolidation goals than their "evolved" counterparts. The reason is that mainframes have been evolving too, adding features customers wanted. They've also been doing things like virtualization and RAS for decades longer.
So, it's hardly evolution: just two similar things competing. I also take issue with the author's conclusion about what going back to the mainframe means. I think we just need to change the goals:
Goal 1: Really inexpensive mainframe-like platform.
Goal 2: Diverse array of companies providing and managing computing resources
Goal 3: Separate (rather that integrate) sensitive stuff, generic stuff, software, storage, etc.
Goal 4: More tech that allows trustworthy use of untrusted components. (eg seemless encrypted data to/from cloud)
These could help us out. The first would let many more organizations have their own easy-to-manage, high uptime, computing service. That defeats many security, privacy and legal worries. The second supports that goal by putting the sea of data into many different hands who have competing interests or goals. That should reduce the risk of collusion or at least the risk of a single large company from exerting undue influence.
Three is already standard in cloud. Examples of three are services for specific functions: compute, storage, crypto, visualization. Decoupling our platforms and applications can allow even more of this in that the components can be deployed in different ways (or locales).
Four is in progress. The best examples of four are TPM- and cryptoprocessor-based designs that protect group A's information on group B's servers. If we keep getting more research done in that area, we can bring the capabilities up and the cost down. This makes it easier to use in these new cloud and mainframe deployments. This boosts our ability to maintain our freedoms in spite of the centralization trend.
So, there's still plenty we can do even though we're reinventing the mainframe. Whether we will remains to be seen.