Re: A common problem with SLAAC in "renumbering" scenarios

[Jan Zorz - Go6 <[e-mail]>]

On 07/02/2019 21:04, Michael Richardson wrote:
> 
> Jan Zorz - Go6 <[e-mail]> wrote:
>      >> This creates two routes per customer.
> 
>      > Ya, I know... Usually this has been handled with creating a pool of IPv6
>      > addresses per each BRAS/BNG where /64-s were allocated for each
>      > connecting CPE to number the WAN link. Experience shows that dynamic
>      > changing of the WAN addressing does not bring much pain, as does
>      > changing of addressing in the networks behind CPE. In this way you have
>      > just one route towards BRAS/BNG for that pool of "WAN link" addresses.
> 
> So half dynamic, and half-static via PD.
> It's not a bad a solution, particularly because it feels more efficient
>  From an address point of view if you have a lot of these single-host
> networks:

This solution saved the day to many ;) However, instead of going around 
the problem - we need to fix it. And that's what we propose in our I-D.

> 
>      > I was surprised to discover how many people still connects to the
>      > internet with one host that does the PPPoE connection (Windows, etc)
>      > and in this case
>      > - not numbering the WAN link would leave this hosts without any IPv6
>      > addresses.
> 
>      > However, to be honest, I never tried what happens if you connect with a
>      > single host PPPoE client that can't do PD and the system on the other end
>      > does just prefix exclude option. Would that host still remain without any
>      > IPv6 addresses? Hmm...
> 
> I don't know either.
> That's why I think we need to single the network scenario at the IP6CP,
> thus the document that we have been discussing in a private thread.

That would be a very useful document. I suggest that we continue that 
thread and get some experienced operators onboard ;) Prague?

> 
>      >> and IPv6 work exactly the same at that point.  They just didn't
>      >> remember that
>      >> IPv4 was hiding the same state in what they think of as layer 2.
> 
>      > Good point ;)
> 
> Also, by "hiding", I also mean: not making the state visible, auditable,
> or easily monitored as to how full it is.  While the ARP cache is a *cache*,
> and can be reconstructed, if it's full, and you have >cache number of
> customers, then each time a customer gets more traffic, a different
> customer goes offline.  It's usually completely invisible to network
> operations.
> 
> Stupid layer-2 tricks that IPv6 eliminates.

Yup, yup ;)

Cheers, Jan