The physical layout of digital infrastructure affects the logical layout in multiple ways: the efficiency and scaling, the co-location of compute to large or sensitive data, the environmental footprint etc.
The eternal pendulum between "thin" and "fat" clients, or "on-premise" versus "cloud" are different aspects of that fundamental conundrum: how should chips be distributed physically?
The answer to that does not seem to be fixed and unique and there are complex feedback loops.
2/
Comparison with the network distribution properties of other utility technologies does not help resolve this either: households may burn their own gas for heating (at least for now), or rely on district heating, or some remote wind farm (or nuclear reactor). Water provision is typically fully centralized in cities but not so in agriculture.
Without proof, intuitively it feels that "hyperscalers" is a bad design for society and environment. Silicon should be as decentralized as realistic...
3/
Which brings to the final point and closing the loop: the feedback effects of software and economic models and organization. E.g., If #selfhosting was dead easy (via NAS etc) it could expand to be as common as an AC plug or water faucet.
Or maybe decentralization could be resolved at municipal level ("city clouds")
Only then can you really look at web protocols and rank them according to fitness.
In the meantime we will have a cacophony of wildly different ideas competing for mindshare
4/4
How about this architecture:
On the client:
- dweb serve: a local web server that has a web API which can be accessed by local browser apps, and any desktop or local client that wants to access...
- #Autonomi - a p2p decentralised storage system that keeps data after a one off payment (no ongoing fees, expiring domains or websites).
dweb is also a client, so 'dweb open <ADDRESS-OR-NAME>' opens a decentralised app or website in your browser from Autonomi via the dweb server.
