The internet derives its strength and flexibility from its design as a decentralised system with the bulk of the inteligence on the edges rather than in the network "core". Unfortunately it is still too centralised in many respects. Much of this centralisation stems from early technological constraints that either no longer apply or will shortly cease to apply. The early internet required central management because it relied on a protocol with a relatively small (32 bit) address space and routers that operate under severe memory constraints.

We can reasonably assume that the number of independent networks will be of roughly the same order of magnitude as the number of people on the planet ie a few billion. Since modern computers come with several gigabytes of RAM we can work on the assumption that storing the routing table is now trivial. Likewise network link speeds are increasing so transmitting the table should not be prohibitive.

What might be expensive is the need to look routes up quickly. This might require very fast RAM on core routers and storing the entire routing table therein would be prohibitive. This could be avoided by making use of a source routing protocol like MPLS to enable the workload to move to the network edge.

Given the above we no longer need routing tables to be efficient. We should be able to afford one table entry per network easily. This means we no longer need management to ensure compact allocations. I could be wrong but I suspect I'm only wrong by a few years if so.

Although allocation compactness is no longer a concern we still can't allocate at random with IPv6 there might be accidental collisions. However we don't need an authority to prevent this just an agreed standard. One mechanism would be to assign each router a network address based on its physical location on the surface of the earth. One could use any map projection that produces a roughly square map without distorting shape or area too badly and simply take the router's cartesian co-ordinates to make up the network address. With a resolution of a square meter this would take up about 50 bits, comfortably within the 64 bits reserved for the network. By interleaving the bits from the X and Y co-ordinates one might even be able to shrink the routing table back down again.

Of course that doesn't prevent hijacking an IP address as there is no central registry of who legitimately controls which address. If one is prepared to throw out IPv6 compatibility and increase the address space then one could just use a hash of the router's public key to identify the network.

Unfortunately Zooko's Triangle causes some problems when trying to decentralise human meaningful names so I'll leave those to a later post.