Re: [v6ops] Objection: draft-ietf-v6ops-ipv6-cpe-router-09.txt

[Tim Chown <[e-mail]>]

On 21 Dec 2010, at 11:39, Rémi Després wrote:

>> I think there's some confusion here.    The MUST is that CE is able to *generate* a randomised ULA prefix as per RFC4193.    ULA-1 isn't about what is advertised.   Indeed the default for whether ULAs are advertised or not is not prescribed in the text. 
> 
> Right, good point.
> 
> The proposal should better have been:
> "ULAL-1:   The IPv6 CE router SHOULD be capable of generating a pseudo-random ULA prefix [RFC4193]. If it isn't, it still MUST have an IPv6 prefix to use when its WAN interface is not operational, e.g. a ULA-like constant values like fd01::/48."
> 
> Note that, since user configuration of ULA prefixes is permitted, defaulting to fd00::/48 is equivalent, in unmanaged CPEs, to having a magic hand configuring it (fd00::/48 is a permitted possible of a pseudo-random generation).

Well, as it was in the -08 version, ULA-1 and ULA-2 taken together allow a CE to have its ULA prefix either automatically or manually configured, though the latter is a SHOULD.    So you don't need your extra text encouraging use of fd01::/48.   Yes, some users may do that, but I don't think we should suggest it.   

The ULA concept isn't really stated anywhere in the text.   Perhaps section 3.2.1 should say something like "Internal IPv6 LAN interfaces may be provisioned concurrently with both a delegated globally unique prefix and a ULA prefix.   The global prefix may then be used for communications to external hosts, while ULAs may be used on the internal network to give persistent addressing through external renumbering events (even for single LAN sites) or in the event of no external connectivity being available".   This may help avoid readers of the draft associating ULAs with NAT66.  

In practice the usefulness of ULAs - at least with current IPv4-style models and without considering SmartGrid type scenarios - will depend on whether IPv6 ISPs give customers dynamic IPv6 prefixes.   While you can argue ULAs may help in the event of a temporary connectivity outage, the existing global prefix should still be sufficient to maintain internal connectivity.    I think these issues are more likely to be incorporated in more detail in the -bis Advanced draft, but we shouldn't preclude the option to provision ULAs in the 'basic' CE definition.

Tim