Byzantium seems to use OLSR. OLSR and BATMAN aren't built to scale, they max at a few hundred nodes per network bc each and every of those nodes needs to store the whole topology of the network and also they need to flood the whole network with reoccurring pings to update the topology and signal liveness.
And they aren't zero-config, you need an "institution" to generate and manage addresses of all the nodes, increasing the maintenance cost. Cjdns just generates a new random address at the initial startup of a node.
Cjdns tried another approach by using DHT routing, that way each node just needs to communicate to and store just a few neighbors. If a node needs more infos it asks its neighbors. But this approach was buggy, maybe it was just a bug in the implementation and not a systemic problem, no one knows. Then they switched to a "supernodes" approach, each SN stores the whole topology of the net, if a node needs to know the route to an other node it asks a SN, there is a "master" SN, all the other SNs connect directly or indirectly to that master SN, they download all the data from the master SN, but that's obviously a lot more centralized than the DHT approach.
Im just saying that we need a lot more investment into that field, just killing the internet wouldn't magically spawn a functioning meshnet (able to replace the internet and handle millions of nodes) in no time.
RE: Anarchy Has Gone Viral - Jeff Berwick at Anarchaportugal