[peterwwillis]

It's not really technological or social, it's logistical.

We've had all kinds of redundant network topologies that used independent networks for decades. The internet is decentralized, and it works pretty well, all things considered. The web is fairly decentralized, too: DNS is independent of a registrar is independent of a network service provider and all are independent of ISP's, and even those are independent of backbones.

The only thing that isn't very decentralized is the client-server IP/TCP/HTTP model. You can provide decentralized versions of HTTP services, but those are the things that are the most costly and inconvenient to decentralize. It can be done, but it's a huge pain with very little benefit.

[hestefisk]

HTTP is client / server (session) based, but TCP / IP certainly isn’t. I’m is act as a client / server protocol because we tell it to, but IP networking was written to be distributed / decentralised so as to provide better resilience, unlike other networking standards at that time (token ring, anyone?).

[peterwwillis]

TCP is connection-oriented, and each connection is a session. IP is decentralized, but the way it's used now in consumer devices makes initiating connections to them difficult and dangerous. Any realistic hope of a successful new distributed web should address this problem, though probably the current solution is "have clients join a private network and route back through it", completely side-stepping firewall concerns. If you ignore the concerns, I guess these protocols aren't that big a stumbling block.