Renumbering an enterprise is hard.
Renumbering an IXP even a large one is relatively simple and has been done
multiple times.
I still don’t support the proposal, but I think that the “renumbering is hard”
argument rings a bit hollow when it comes to IXPs.
The process boils down to:
1. send out notice and map of old addresses to new addresses.
2. Add new addresses to route servers (if applicable)
3. Give time for providers to bring up all the new peering
sessions on the new addresses (in parallel to the existing ones)
4. Turn off the peering sessions with the old addresses on the
route servers (if applicable)
5. Set a flag day for turning off peering sessions on old
addresses.
6. Filter TCP/179 traffic to old addresses on flag day. (let other
traffic continue to pass)
7. Wait for traffic from deprecated peering sessions to drain off
8. Set a flag day to deprecate the old addresses
9. Filter the old addresses on the flag day.
Is it pain free? Not entirely, but relatively so. Main source of pain is
providers that have to scramble after they
ignore the flag day notices.
Is it particularly complicated or difficult? No, not really and if providers
are paying attention and cooperate before
the flag day, it can be non-disruptive and relatively pain free.
Owen
> On Aug 12, 2023, at 00:37, Christopher H <[email protected]> wrote:
>
> Hello Team,
>
> I support parts of this proposal, while I oppose others.
>
> In some economies (to use Australia as an example), there are significant
> numbers of network operators. If an IXP were to start out and then have a
> requirement to re-number and expand, the bigger the IX becomes the harder it
> becomes to renumber. Let's look at MegaIX Sydney, and hypothesise that this
> policy was in place when they were reaching 80% utilisation (204 IP
> addresses) of a /24 subnet. It would be a significant challenge for all 204
> peers, plus the route servers, to renumber and re-establish their peering
> sessions. The more peers you have, the harder it becomes.
>
> Let's look at other economies such as Vanuatu, where they have such a small
> IX. I feel that in circumstances like this, it's not justifiable to allocate
> an entire /24 to an IX which has less than 5 peers. Given the size of the
> economy, it's unlikely (for the foreseeable future) that they will see
> requirements for anything greater than a /26 subnet.
>
> Having said the above, we cannot discriminate economies based on the number
> of participants in delegating assignments. It may be better suited to
> restrict delegations to a /24 then if the need arises to renumber to a /23
> they are given a 6-month window to return the former holding, and if they
> need to go from a /22 they are given 12 months. Whilst they hold these
> resources during the transition, they are responsible for any membership fees.
>
> Regards,
> Christopher H.
> _______________________________________________
> SIG-policy - https://mailman.apnic.net/[email protected]/
> To unsubscribe send an email to [email protected]
_______________________________________________
SIG-policy - https://mailman.apnic.net/[email protected]/
To unsubscribe send an email to [email protected]
