It's not hard to achieve separation in cloud environments. It's just expensive, and for most, the need doesn't justify the expense. Every once in a while you do run into things like actually all the clouds in (location) are located in the same building; or you paid for two separate strands, but a backhoe has determined that both strands are in the same bundle. Or your two redundant providers of whatever use each other to provide service.
With more complexity comes more interesting failure scenarios. Simple redundancy handles problems where components outright fail, but software never really ... fails, in that sense. It always performs exactly to design – but sometimes the design is inadequate. Replication and majority voting does not handle design flaws and errors of e.g. interaction, which are increasingly common the more softwareified a product is.
Thus we need more complicated ways to build fault tolerance into systems where we have to assume that the design is wrong somewhere, and we cannot just replicate ourselves out of the problem.
One detail that caught my attention: they used triple-redundant IBM 7090 mainframes with real-time voting between systems. This is surprisingly similar to modern consensus protocols like Raft, just implemented in hardware. Makes me wonder if we've over-complicated things with software-based consensus when simple majority voting could suffice for many use cases. The key seems to be truly independent failure domains - something NORAD achieved through physical separation that's harder to replicate in cloud environments.
Would love to hear from others who've implemented hardware-based consensus systems. How do they compare to software solutions in terms of reliability and complexity?