The "punch out" part is the problem. You can't "punch out" without using special tricks (send UDP source port 1234 to target port 1234 on both machines at the s

The "punch out" part is the problem. You can't "punch out" without using special tricks (send UDP source port 1234 to target port 1234 on both machines at the same time and then pretend that UDP is an established connection to get through your stateful firewall) If your firewall randomises your source ports then sure, you have to use the birthday problem style tricks that tailscale uses, it's not onerous though. > Aren't these features of your router, not of your IP stack? Yes, and that's where I want them to stay. Which means NATing depending on which direction I want to send the traffic (and get return traffic) -- even if I have a BGP handoff upstream. So ipv6 doesn't get rid of NAT's use, just changes it to a 1:1 mapping which is a minor benefit (and renames it to NPT) > This sounds like a pretty niche application, but sure. The internet has two sets of people 1) Consumers who just want to establish https connections to server, in which case they don't care about NAT, CGNAT, etc 2) People with niche applications NAT is a very useful tool, and the ipv6 fanboys that go on about how evil it is just want to take that ability away from people because they don't understand it. Most of the arguments against NAT stem from a time when stateful firewalls were not a thing. > Right now you're using the NAT as a kind of forwarder to send the data to different hosts, so if you have a router you can run software on you could just have it forward to both devices on the local network. Yes, this software runs at a layer 4 level and forwards the selected traffic by translating the address. That's exactly what NAT is, it's great. I'd be quite happy running an ipv6 network with network translation but given that far too many things simply don't work on an ipv6 only network (tv, nintendo switch, zscaler laptop), and those that do require 64 translation (github) IPv4 addresses are not increasing in cost by the way - in nominal terms let alone adjusted for inflation. In real terms they're 20% cheaper than 2019 https://auctions.ipv4.global/prior-sales https://auctions.ipv4.global/prior-sales