From this thread: https://cpunks.org//pipermail/cypherpunks/2014-July/004957.html

On Sun, Jul 6, 2014 at 3:27 PM, Lodewijk andré de la porte <l@odewijk.nl> wrote:

2014-07-06 19:42 GMT+02:00 grarpamp <grarpamp@gmail.com>:

...

IPv6 To be honest, too centralized

There's nothing centralized about it at all. You generate your own rfc4193 IPv6/48. It's random so still check with other projects to be 100% sure before going live, and there's even an open registry to help with that. The problem is when people like cjdns blindly stake the other whole available even wider fc00::/8 as their own without thinking about interop (well if you actually need the bits that is, but with accepted crypto starting at 128 / 2048, 120 bits is no more valid width than 80. and with the DHT below this native width limitation is moot).

and too arbitrary. It's odd to declare IP(v6) addresses bullocks that simply. It's in our systems and in our thinking.

The reason to declare IPv6 on a tun interface on your host is so users can run whatever IPv6 enabled app users want to run, on whatever darknet, right now, today. Darknet projects should be coding good darknets, not duplicating existing writing user facing apps for them (browser's, mua's, login's, fileshare's, git's, wiki's, webserver's, etc).

But really, universally unique addresses are just a dream. The only address I think makes sense is derived off a public key. Any other address should be network-topology-semantic, not assigned by committee. That way there's real identity (private-publickey) and real address/location (place dependent or similar address)

An IPv6/48 is big enough to hold every user on every darknet simultaneously. But yes, we cannot arbitrate control central the map from IPv6 addr to user (truly, their darknet internal wide address / PKI key). So DHT... darknet user picks random IPv6 addr as their own key, adds in each of their darknet addresses with sigs over it [1] as values. Now you have the needed interop map. You need to label and rewrite packets to/from each darknet with this scheme, but that is easy part. [1] To be flexible in binding, one darknet per port/range, or one per whole IP. The DHT could hosted over any darknet, or it's own special darknet.

One of these months I will continue work on a paper reg. semantic addressing and a true IP layer replacement. I feel bad having talked about it for years now, but life and inexperience haven't really allowed me to make a great paper out of it. One of these months.

The problem with IP stack replacements to something really wide (eg: 512) is you then have to produce a library and beg all the above user apps to link it and handle it... not going to happen. Or write your own apps which takes the whole variety of apps/life on clearnet and crushes it into your few apps... not fun. So replacement = limitation and poor adoption.