distributed system relies on the central DNS hierarchy

[anarcat]

one flaw in the distributed system we have is what happens if a city loses the uplink to the internet. this could happen in russia, if russia decides to finally get off the global network, or in ukraine if a city gets cut off from the world.

in this case, names like matrix.kiyv.dcomms.net.au won't work, and users will fail to discover the nodes and possibly also the rooms. the matrix.to links would fail as well.

a workaround to this problem would be to run a DNS server on the nodes and somehow make that discoverable in the city. running bind or some authoritative DNS server on the node could allow a delegation from the dcomms.net.au zone to the specific city and then, as long as the glue records are available, DNS would still work.

the problem is when those glue records fail as well... at that point you're kind of screwed and can only rely on IP-level routing. In that case it might be useful to have the hosts answer on the IP directly, as a vhost, but server-to-server communications would possibly fail as well anyways.

so local discovery. that could be done with the bonjour/rendez-vous protocolo, but that assumes a local LAN. i would need to know more about the network topologies to make this work, but needless to say that if a city is completely cut off from the network anyways, i guess there are many other problems to worry about...

also consider that there are p2p efforts in Matrix which could help bypass that problem as well, see the pinecone project, but none of that is production ready.