On 5/16/25 17:30, Tim Burke via NANOG wrote:
Trying to do hacky things with CGNAT to save a buck is, IMHO, inexcusable, especially when lots of FTTP operators are now overbuilding legacy ILECs/cablecos with fiber that is typically being promoted as “superior in every way”.
If customers were willing to pay for it, they'd be more likely to get it. Unfortunately, getting a customer to pay more than what the incumbent LEC/MSO charges for legacy DSL/DOCSIS service is tough, and the only part of the equation that a new greenfield fiber carrier can compete on is speed since it's effectively unlimited for them. I've taken to putting residential customers behind statically-mapped 16:1 or 32:1 CGNAT444 (with native, hardware-forwarded IPv6) by default and then just moving them to 1:1 public space upon request or for any form of repeated trouble calls that seem like they may be related to NAT in one way or another. That drastically cuts down the number of addresses necessary while keeping almost everybody (including customer support on my end) reasonably happy. I'm trying very hard to get IPv4aaS-over-IPv6 usable so that I can make things even simpler and more transparent for my users. Sadly this has not taken off nearly as quickly as I would have liked aside from 464XLAT which really doesn't solve the problem I care to solve (in fact, it arguably makes it worse). I've also sadly still seen far too often CPEs and public Internet endpoints neglecting IPv6 to the extent that it performs noticeably worse than IPv4 even when the observed AS-paths are identical. This definitely does not help matters as it tends to drive end users to disable that native IPv6 that I do provide.