Re: RS sending in draft-ietf-v6ops-ipv6-cpe-router-04

[Ole Troan <ot@cisco.com>]

Philip,

>>> Some ISP do not assign any address to the WAN link and the DHCP-DP RFC prev
>> ents
>>> me from assigning a address from the prefix to that link. That annoys me to
>> no
>>> end.
>> 
>> I don't understand what the problem with this is. the CPE draft
>> recommends that an address is configured on an internal interface.
>> why does it matter to which interface the address is on?
> 
> Because, with picking a global address from a random interface it is mich
> harder to ensure the ICMP error gets te source address I want it to have.

who are you?
for a an ICMP message sent by the CPE to some remote destination I can hardly see it matters.

> And also because the reason for disallowing it is not all that strong.
> 
>>> But worse, when every ISP comes up with an arbitrary list of features (and
>>> requirements) it becomes almost impossible to help a random person with his
>>> internet connection.  Oh, you don't have a global IP address on WAN link, i
>> s
>>> that normal because the ISP doesn't assign one, or did something go wrong?
>>> Neighbor discovery doesn't work, is that normal because the ISP disabled it
>> ,
>>> or is something really broken. You only know if you know the details of
>>> every ISP.
>> 
>> what arbitrary features are those? there are 3 addressing models.
>> SLAAC, DHCPv6 and "unnumbered".  all models use SLAAC.  ND is used.
>> I'm not aware of anything that requires 'details of every ISP' to
>> understand.
> 
> If you come across a CPE that fails in some unspecified way (somebody asked you
> for help because the Internet is down or something like that) how do know
> what is normal or not?

I would love to be able to tell SPs exactly how they should operate their networks. perhaps luckily for them I can't.

> Not all CPE have extensive debug interfaces where you dump all packets.

I would expect them to be able to show you the contents of received RAs though.

> If all links have global IP addresses then you know what is going on if you
> find one that doesn't.

but that's not quite what you are advocating. an address from the delegated prefix is may be configured on the interface, but the prefix certainly wouldn't be on-link.

> I don't know about the 'ND is used' part. But my ISP's router does not respond
> to NS messages (on a PPP link). I think they believe that that's allowed by
> the standards. Are you saying it isn't?

no. e.g NUD should be supported on any link, including PPP links.

>>> But more practically speaking, how do you know the global IPv6 address of
>>> your customer? With SLAAC you would have to know the MAC address to even
>>> guess that address. Do you guess it based on the link-local address? Do you
>>> call the customer? What if the customers CPE uses temporary (privacy)
>>> addresses?
>> 
>> if you are an SP that requires to know the customers address, then
>> you use DHCPv6 for address assignment.
> 
> 
>>> With prefix delegation, a router is very likely to have at least one global
>>> address from the delegated prefix.
>> 
>> yes. are you arguing that this address should be well known? since
>> there is no way you can assign that via DHCP. or are you arguing
>> that there should be a mechanism to discover it? e.g by using the
>> subnet-router anycast address?
> 
> If people want to ping routers that get a prefix using PD then that seems to
> be a sensible approach. 

I suppose we are digressing here... my apologies.

cheers,
Ole