5 ms·

Can you clarify what you mean by multi-link routing? Are you referring to multipathing? What is stopping you from using multiple connections to the internet?

by ra1n85 9y ago

gerdesj 9y ago
Say you want multiple links to the internet - redundancy/cost/whatever. Each link will have a prefix that you are delegated and your SLAAC or DHCPv6 will give out addresses etc. Now which one will your device use? Your PC cannot know that link C is down because that is only known by your router. IPv6 does not address the same basics that IPv4 failed at - routing around crap, without a routing protocol.
- tmiller12 9y ago
  You can purchase your own IPv6 block from an RIR and do multi homing.
  - gerdesj 9y ago
    Yes you can and it will: * add another entry to the global routing table thus nullifying a benefit of IPv6 * cost a bloody fortune (approx £5,000 per year) * is your mum able to ask for BGP peering I am personally able to pay the price for PI but not everyone is and that is my fucking point. Why should your mum not be able to ask for a internet connection that is able to tolerate a failure? There is NPT but it is simply NAT via another name. I've been running IPv6 at home for quite a long time. I passed the WAF at least two years ago. Any more ideas? Cheers Jon
    - buzer 9y ago
      I haven't tried it myself, but I guess one option would be to use DHCPv6 with low lifetime timers? That way router could stop advertising the prefix from upstream that no longer works.
- zAy0LfpBZLC8mAC 9y ago
  > Your PC cannot know that link C is down because that is only known by your router. And your router should withdraw the prefix from RAs, so the PC stops using it.
- ra1n85 9y ago
  That makes no sense. The kernel routing table of your PC should remove the link once the PHY has been detected to be down. That's even if your PC supports multipathing. We don't rely on layer 3 to communicate a link is down to us - we rely on layer 1.
  - izacus 9y ago
    When cable/DSL falls down, it's not going to take PHY down. You'll have active ethernet but no packets will come through. With IPv4 the router will just fix up NAT. With IPv6 you now have to deal with significantly more complexity.
    - ra1n85 9y ago
      >With IPv4 the router will just fix up NAT. With IPv6 you now have to deal with significantly more complexity. How so? In a multihomed environment, you'd still be PAT'ing to unique addresses on the WAN side (unless a FHRP is in use, which would be rare) - if one of those links go down, so too does all of the flows associated with address of the link. If no dynamic routing protocol or redirects are in use, the gateway with the down link will also eat half (assuming two gateways) of all egress traffic. IPv6 should be easier as no NAT/PAT is in use, therefore you have no challenges with statefulness or routing symmetry.
      - izacus 9y ago
        It SHOULD be easier. It's not on current widely available prosumer hardware. Yes, your connections will get temporary eaten. But for the typical developers workload that won't be a huge issue because most connections you do are short-lived. Configuring all your clients to properly switch vs. configuring your NAT router (with builtin functionality for exactly that) is orders of magnitude more complex.

Can you clarify what you mean by multi-link routing? Are you referring to multipathing? What is stopping you from using multiple connections to the internet?

by ra1n85 9y ago

Can you clarify what you mean by multi-link routing? Are you referring to multipathing? What is stopping you from using multiple connections to the internet?

gerdesj 9y ago

Say you want multiple links to the internet - redundancy/cost/whatever. Each link will have a prefix that you are delegated and your SLAAC or DHCPv6 will give out addresses etc. Now which one will your device use? Your PC cannot know that link C is down because that is only known by your router. IPv6 does not address the same basics that IPv4 failed at - routing around crap, without a routing protocol.

tmiller12 9y ago
You can purchase your own IPv6 block from an RIR and do multi homing.
- gerdesj 9y ago
  Yes you can and it will: * add another entry to the global routing table thus nullifying a benefit of IPv6 * cost a bloody fortune (approx £5,000 per year) * is your mum able to ask for BGP peering I am personally able to pay the price for PI but not everyone is and that is my fucking point. Why should your mum not be able to ask for a internet connection that is able to tolerate a failure? There is NPT but it is simply NAT via another name. I've been running IPv6 at home for quite a long time. I passed the WAF at least two years ago. Any more ideas? Cheers Jon
  - buzer 9y ago
    I haven't tried it myself, but I guess one option would be to use DHCPv6 with low lifetime timers? That way router could stop advertising the prefix from upstream that no longer works.
zAy0LfpBZLC8mAC 9y ago
> Your PC cannot know that link C is down because that is only known by your router. And your router should withdraw the prefix from RAs, so the PC stops using it.
ra1n85 9y ago
That makes no sense. The kernel routing table of your PC should remove the link once the PHY has been detected to be down. That's even if your PC supports multipathing. We don't rely on layer 3 to communicate a link is down to us - we rely on layer 1.
- izacus 9y ago
  When cable/DSL falls down, it's not going to take PHY down. You'll have active ethernet but no packets will come through. With IPv4 the router will just fix up NAT. With IPv6 you now have to deal with significantly more complexity.
  - ra1n85 9y ago
    >With IPv4 the router will just fix up NAT. With IPv6 you now have to deal with significantly more complexity. How so? In a multihomed environment, you'd still be PAT'ing to unique addresses on the WAN side (unless a FHRP is in use, which would be rare) - if one of those links go down, so too does all of the flows associated with address of the link. If no dynamic routing protocol or redirects are in use, the gateway with the down link will also eat half (assuming two gateways) of all egress traffic. IPv6 should be easier as no NAT/PAT is in use, therefore you have no challenges with statefulness or routing symmetry.
    - izacus 9y ago
      It SHOULD be easier. It's not on current widely available prosumer hardware. Yes, your connections will get temporary eaten. But for the typical developers workload that won't be a huge issue because most connections you do are short-lived. Configuring all your clients to properly switch vs. configuring your NAT router (with builtin functionality for exactly that) is orders of magnitude more complex.