[simonh]

I think Unix's refusal to impose arbitrary structure on data is what has enabled a lot of the innovation and experimentation with alternative ways of organising and accessing data. Binary blob, SQL, NoSQL, key/value, column oriented, etc, etc Unix doesn't care. It just gets out of your way and lets you do your thing.

A lot of projects tried to implement opinionated OSes to the point that the OS was really an application. I'm thinking of Sprite and Amoeba, which attempted to abstract out the network and present a unified interface to accessing resources, and fancy storage systems, but in doing so impose opinionated interfaces on that access.

That's half way between an OS and an application like Hadoop, or Kafka, but you couldn't effectively develop something like Hadoop or Kafka, with your own optimisations and implementation tradeoffs on one of these things because the implementations or access models would clash.