I’ll be honest: IPv6 has always been one of those things I knew I should understand better, but every time I sat down to learn it, my eyes would glaze over. All those colons, the huge addresses, talk of "prefixes" and "ULA" and "link-local" — it felt like learning a whole new language. And IPv4 works fine, right? So I’d close the tab and go do something else.
But recently I got my home network sorted — running a physical host, several virtual machines on Hyper-V, and multiple services on each VM — and I kept hitting the same friction with IPv4: NAT headaches, conflicting ranges, remembering which static IP belonged to what. I figured it was finally time to bite the bullet. Here’s what I’ve worked out — and if you spot anything wrong, please tell me!
What I started with
First I logged into my router and found exactly what it was giving me:
| Detail | Value |
|---|---|
| WAN Global Unicast Address | 2a01:23c3:33c8:3400::1 |
| Delegated prefix | 2a01:23c3:33c8:3400::/56 |
| Router LAN side address | 2a01:23c3:33c8:3401:e675:dcff:fea5:3ada/64 |
| Link-local address | fe80::e675:dcff:fea5:3ada/64 |
That /56 means my ISP has given me a block of addresses — and crucially, I can split that block up into smaller pieces for my own use.
The big "aha" moment
This is the bit that finally clicked for me: in IPv6, every separate network gets its own /64 subnet. That’s it. That’s the rule I was overcomplicating. You don’t carve it up into weird little slices like you sometimes had to do in IPv4 — you just give each logical network its own full /64 and walk away.
And each /64? It has space for 18 quintillion addresses. I don’t know about you, but I think I can make do with that. I will literally never run out. It feels decadent.
So I decided I only actually need three networks in my house:
| Network | What it's for | My assigned Global prefix | My ULA prefix |
|---|---|---|---|
| Physical LAN | My actual PC, phones, TV, printer — anything plugged straight into the router | 2a01:23c3:33c8:3401::/64 |
fdda:19e0:19c5:1::/64 |
| VM Network | All my virtual machines, connected through the Hyper-V internal switch | 2a01:23c3:33c8:3402::/64 |
fdda:19e0:19c5:2::/64 |
| Services Overlay | The extra virtual adapters inside each VM — one per service so each has its own unique address | 2a01:23c3:33c8:3403::/64 |
fdda:19e0:19c5:3::/64 |
That’s the entire plan. Three lines. That’s all I need.
The private addresses I didn't know I needed
There was one extra thing: ULA addresses. Think of these like the 192.168.x.x or 10.x.x.x ranges you already know — private addresses that work only inside your house. They never change, even if your ISP decides to give you a different global prefix tomorrow.
I went into my router settings, flicked the ULA toggle on, and got myself a private range: fdda:19e0:19c5::/64. From there I just assigned a different chunk to each of my three networks by changing that 4th number — easy.
Now every single interface gets two addresses:
- A global one for talking to the internet
- A ULA one for talking to everything else in the house
It’s surprisingly liberating knowing my internal addresses will never break just because my ISP did something weird.
How it all fits together
- My physical PC lives in both the physical LAN (
::10) and has a virtual adapter sitting in the VM network (::1) — that way it acts as the bridge between them - Each VM gets one address in the VM network for its main interface
- Each service adapter gets one address in the services subnet — no gateway needed, they just talk locally
- I only ever set one default gateway on any machine — on its main physical or VM interface
No NAT, no port-forwarding gymnastics between internal hosts, no guessing which IP maps to what machine. Just clean, straightforward addressing.
Things I'm still not 100% sure about
This is where I'd love your input — am I getting these bits right?
- ✅ Using the router's link-local address as the default gateway on devices — it works, but is that the "proper" way? It means if my global prefix changes, nothing inside my house needs reconfiguring
- ✅ Only putting a default gateway on the primary interface — all the secondary service adapters sit quietly in their own subnet without one. That feels right, but I want to be sure
- ✅ My host has an interface in both physical and VM subnets and has IPv6 forwarding turned on — is that the cleanest way to get traffic between them?
It feels like I’ve only scratched the surface, but honestly? I was expecting this to be ten times harder than it actually was. The biggest barrier was just my own hesitation. Once I accepted "one /64 per network" and stopped trying to make it look like IPv4, everything fell into place.
sets reminder to read this again on Monday as I'll probably forget it again over the weekend! 😀
Top comments (2)
The point about one /64 per network is the clearest part of this write-up, because it replaces the IPv4 habit of carving up scarce space. Using the router link-local address as the default gateway is also the usual practice, since link-local addresses stay stable when the global prefix changes, and that stability is exactly why the ULA addresses help here. The one detail worth watching is the host acting as a bridge between the physical LAN and the VM network with forwarding on. That works, but it makes the host a routing hop, so its firewall rules and forwarding settings deserve the same care as the router settings.
When you've dealt with so much scarcity it really does feel wasteful to have a /64 for a home network! Having grown up with home computers like the BBC Micro and ZX Spectrum I still have that mindset of trying to get the most out of everything, it's a relief but also rubs against the grain a bit to have 18 quintillion addresses available that i'll never use 😂
Great point about the bridging, I want to avoid it as much as possible because it seems to slow everything down - especially with Hyper-V. One thing I want to test is whether a pure IPv6 setup will be faster than IPv4, i'm still running a dual setup for now.