Can you describe the routing problem a bit more in detail? I'm trying to understand why multi link should be more a problem in IPv6 than in IPv4.

If you are multi homed it sounds as if you should have your own /48 IPv6 network (PI) and then only use the link addresses for routing - but maybe there is something that prevents you from - looking forward to hear your fully story!

Multi-homed doesn’t help when you have two links to two different ISPs that you want to load balance. You need NAT for that. Also, I haven’t looked at multi-homing costs but I assume they are not free (let me know if I’m wrong). Plus in a small office or retail location, you don’t need a consistent public IP when you aren’t hosting servers (or they are only via VPN).

What. If your ISP is not terrible, it just takes a route advertisement to switch the same IP address between networks. And route the same subnet through multiple networks. There is no need for OSPF, RA is good enough. The ISP just has to not filter them out and forward into routing. (You will have to inform them that you use a specific separate subnet on your network. This means typically a business contract.)

Not all ISPs offer this functionality. I can’t realistically tell the customers I manage that we can’t support their preferred provider because of this, when they rightly point out that it just works on IPv4 without need for multi-homing.

And again, multi-homing does not address load balancing / policy based routing with active active links. Eg it’s possible to have 2 Ethernet connections, 1 VDSL connection and a LTE link. Can multi-homing IPv6 work in such a situation while providing fast failover and load balancing by service/application/policy?

I wonder as well. The main downside I have heard about IpV6 until now is that many softwares are not compatible.

I have to admit there is actually some. Luckily though, when we saw problems, recent updates (i.e. mosh, ping, etc.) actually fixed them.

However there is some really broken software out there that hard codes IPv4 addresses - which even breaks NAT64/DNS64!

Which for example?

Oh, and not to forget: the ever broken cpanel.

Video games for example.

Can be fixed

mosh was such a candidate, but fortunately has been fixed

I've got the exact same issue.

The solution when I investigated was to use IPv6 network prefix translation; however most routers do not support it (including Mikrotik routerOS, which is what I mostly use).

So lacking that, instead I've gotten my own AS and am now a 'proper' internet peer.

So... you have multiple advertising routes, and each box has 5 public addresses, and the complaint is that the boxes can't magically decide how to handle all of the different routes?

When it comes to routing, you have two options, delegate or self-manage. In the case of self-management, yes, if you want complex route management, then you'll need software that handles complex route management.

Why not use Stateless DHCP, and configure a machine with short leases that can update the others as needed? It can handle the complex routing, and the clients just have to frequently ask if there are updates.

In my case, I want to fail over from VDSL to 4G within a couple seconds of VDSL going down. Will short leases cover that?

I also like to load balance traffic on another site with two VDSL links and a 4G backup. I can’t do that without NAT.

DHCP traffic is approximately 512 bytes in each direction for each half lease period. Stick 1000 clients on a single node, with 2 second leases, and you'll use 0.5% of your bandwidth, and be able to update your default gateway every second

Leases? What leases, this is not DHCP. If you mean Route Advertisement with short life, then yes.

The leases referenced by the post I was replying to:

“Why not use Stateless DHCP, and configure a machine with short leases”

How do you deal with one missed RA causing an inadvertent failover to using a different WAN, while maintaining 1-2s failover after the first WAN link goes down?

Also this does not fix the issue with inability to load balance without NAT. Eg I have a desktop (or tablet etc) that I want to send VoIP out one WAN and everything else out another.

Sounds like you should have your own PI space if you have multiple uplinks, and have that PI space advertised out across all four links. Then you can route packets however/wherever you want using whatever policy you'd like.

This is cost prohibitive, and I doubt most residential ISP's have the capability for this. I would like to have both a cable and a DSL ISP for redundancy but it's far from straightforward with IPv6 (and I have run IPv6 networks for not far off 20 years).

You are so far outside the norm here... Do prefix translation, it's like NAT, but run ULA space inside.

Now you can load balance across all your links without issues.

How is this particular situation better with ipv4?

In IPv4 people have accepted that NAT is the best they'll ever get and it's pretty easy to do redneck multihoming by round-robin NATing a private network behind multiple public IPs. In contrast, some people thought IPv6 was going to violate the laws of physics in their favor so they're disappointed that it hasn't happened.

BGP, most likely.