Comment by lazide

Oh boy.

You don’t need to allow peer-to-peer connections with IPv6. They’re easier to allow and book keep - but also easier to block. The workarounds for peer-to-peer with IPv4 NAT are extremely difficult to detect and stop (STUN, various proxying setups, etc.). A lot of software does it though, for performance reasons. CGNAT is quite expensive and error prone, and causes a lot of support calls too.

Every ISP router I’ve gotten (US, India, Brazil, Germany) in the last few years had IPv6 AND default block for inbound connections in the stateful firewall. Which is fast, cheap, and easy. And most of my traffic (~90%) ended up being over IPv6 by default in a dual stack environment, with certainly no apparent latency penalty. In most situations, a latency decrease near as I can tell, as I didn’t need to route through someone else’s random servers at first to initiate connections for certain kinds of traffic. And no, I wasn’t torrenting.

The hilarious thing here is what is even the fight about?

There are too many humans on this planet for even one IPv4 address per, and too much traffic/connections to sanely coalesce every thing under CGNAT - and why go through all the trouble, when IPv6 is simpler and faster at an infrastructure level anyway than multiple layers of CGNAT and dealing with all the crazy BS that comes up when you have that much address translation and packet rewriting going on.

Which, notably, is more expensive than the more straightforward stateful firewall stuff anyway.

No one is intentionally going to IPv4 unless they have no choice due to backwards compatibility, and that is an increasingly shrinking pie. In another 5-10 years as the old consumer gear finally EOLs, it’s probably going to only be used for niche backwards compatibility (like RJ11 and the old school telephone system), and corporate use where their EOL timelines look more like 50 years. But pipe over tunnels over IPv6.

Which works great BTW - 90% of my active IPv4 usage is for internal servers using Tailscale, which is all actually transported over IPv6. And it does that because while it can use CGNAT punching tricks with TUN/S, etc. it’s faster to just connect directly (through the firewall rule I explicitly created to allow this).

And that is just because the Tailscale software prefers to display/default copy-paste it’s internal IPv4 addresses over internal IPv6 addresses for some reason, which I’m sure will change at some point.