Selfhosted

This is an old post about ipv6, but it inspired me to go looking, and I wanted to share my findings.

1. for globally routeable IPv6 addresses, probably do let it happen automatically, either direct from the ISP, through the router by prefix delegation, or your own implementation of prefix delegation.

2. for devices you want to access, internally, create a ULA within the fd00::/8 space, and assign numbers (and names) however you like. Translate all your 192.168.x.y IPv4 addresses to fd00::x:y and go. Only limitation is you won't be able to access those devices, using the ULA, from outside your network.

3. you can do both of these on the same subnet, and devices pick up both addresses then use the global address for internet and the ULA for intranet.

That means you can do dhcp, dynamic DNS, private domains, and all the stuff you know about IPv4 for IPv6, and still do all the stateless autoconfig that "they" want. Some devices, like my android phone, never played well with dhcpd6, but immediately preferred IPv6 as soon as I let them SLAAC.

If the prefix assigned by the ISP doesn't change, then device SLAAC address shouldn't change, either, because they're calculated from MAC, so if you need to access some internal devices from the internet, you have to mark that address, but (IMO) marking the full address is not that much worse than marking the prefix and remembering the device number.

Ipv6 requires fundamental rethinking about how addressing is done. If you're trying to apply v4 concepts to V6 you likely end up running into something they intentionally designed out.

A unique local address is an address space where you could do that. It's the equivalent to RFC1918 eg. 172/192/10. So you could statically assign fd0::x, and that is expected, but not required generally.

I wouldn't give each device a static unique global address unless they need to be accessed via wan without domain consistently. You lose device privacy really quickly that way because every device gets a unique globally routable address. It's fine for internet facing services but most Linux, Windows, and mobile implementations are using ipv6 privacy extensions by default to ensure you get a random GUA every day.

My network is dual stack and I connect mostly over ipv6 to all my internal clients using internal DNS. If my internal DNS is ever down I can fall back to ipv4 or it's basically the one box on my network with an easy to remember ULA.

On one hand you definitely don't want to be assigning manual/static IPv6 to all your devices because if your prefix ever changes you'll have to update it everywhere. IPv6 doesn't really have a concept of private address space (with a few exceptions). ~~On the other hand most modern IPv6 stacks support dynamic protocols like SLAAC while also assigning a static suffix to the published prefix (e.g. You want :0:0:1234:1 to go to your server, and SLAAC gets the prefix 200x::5678/64 your server would assign itself 200x::5678:0:0:1234:1).~~

DHCPv6 fixes a lot of these headaches for managed networks by allowing you to reserve specific IPv6 for a given DUID.

IMO, your network, do what you want. ~~I have two jump Raspberry PIs that I have static suffixes so I always know where they are without relying on DNS or whatever.~~ Edit: I apparently misremembered how I had these setup. I use a custom interface up script to take the SLAAC prefix and append the custom suffix to it as a secondary IP.