18

What are the differences between multihoming in IPv4 versus IPv6?

Can an enterprise request a provider-independent IPv6 address space from its RIR/LIR which can be advertised to multiple upstream providers via BGP just as in IPv4?

Are the rules for requesting a provider-independent IPv6 allocation the same for all RIRs?

14

You can indeed request a Provider Independent (PI) assignment from the local RIR through an LIR. Routing a block of IPv6 address space is done with BGP in the same way as a block of IPv4 address space. The block is only a bit bigger :-)

For IPv4 it is common that a block smaller than a /24 (so a /25 or longer prefix) will not be commonly routed by ISPs. In IPv6 the common limit these days seems to be a /48.

Every RIR has its own policies, so you will have to look at the RIR in your own region to get the specifics. If you happen to be in the RIPE NCC service region I can answer any questions you might have.

  • For enterprises with sites in multiple regions is it generally advisable to request separate allocations from each local RIR or to use a larger allocation from one of the RIRs to cover the enterprise? – User123456 Aug 31 '13 at 16:04
  • I think the general rationale for allocation seems to be "Keep it tidy" therefore a larger allocation (such as /40 or /32) would be more welcome than several, spread around allocations all tied to the same Organization. – ItsGC Aug 31 '13 at 16:15
  • If you are going to announce a separate /48 for each location into the routing table then it doesn't really matter globally if they are consecutive or not. It might make them easier to remember though. – Sander Steffann Sep 1 '13 at 16:03
4

This seems to be the approach that a number of businesses are trying, but the goal behind the IPv6 design was to prevent all but peering-sized companies (e.g., Google) from getting provider-independent blocks to reduce the size of the global routing table.

IPv6 hosts are required to be able to handle multiple addresses per interface, and the intention was for multihoming to work by having the enterprise's egress routers each advertise the block (generally a /48 or /56) available via its uplink, and for the routers inside the enterprise to append the global prefix (generally read via DHCPv6) to a prefix-independent subnet number. Migrating hosts that are getting their information from router advertisements can be done gradually and without admin intervention.

Unfortunately, in actual deployment this model was hampered by the adoption of the AAAA DNS record (which stores just a literal IP address) over the A6 record, which allowed specifying address components (e.g., an enterprise-wide 48-bit prefix part and an 80-bit host part) that could be managed and update independently; and by flaky prefix-based address support in early router software versions, and it seems rather unlikely that the multi-address model will gain any traction over the PI+BGP model. Early RFCs recommended against assigning PI blocks to non-transit organizations, but as of at least RFC6177 this recommendation seems to have been withdrawn.

0

The original idea of the IPv6 proponents was that organisations would run multiple address blocks in parallel to allow multi-homing.

However in practice this is problematic for several reasons.

  1. In such a setup when an end-host chooses a source IP it is essentially making routing decisions but end-hosts are poorly placed to make routing decisions.
  2. Adding or removing IP prefixes is often difficult as IPs are stored in many places. There were proposed extensions to DNS to help with this but they added a load of complexity and fragility to the DNS system and were eventually abandoned as "historic".
  3. It doesn't really handle the case when connectivity to one provider unexpectedly goes down.
  4. Routers will need to make routing decisions based on the source IP. Some routers can do this but it's an advanced feature, not normal routing.

Eventually it seems the powers that be realised that organisations weren't going to put up with that shit and if they wanted to see IPv6 adopted they had to offer IPv6 PI space on similar terms to IPv4 PI space.

Afaict exact policies vary a bit between RIRs but generally if you can demonstrate intent to multihome you should be able to get a block of PI space without too much difficulty.

It will be interesting to see how this plays out in the long term. With IPv6 NAT being strongly discouraged I can see an explosion in the routing table size as medium sized enterprises move from natted v4 to provider-independent IPv6.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.