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

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

Richard,

On 05/02/2019 20:00, Richard Patterson wrote:
>      > Why?
>      >
> 
>     Here I'll have to repeat what Ole said at the beginning of this
>     conversation. If you setup a lifetime of prefix delegation and
>     assign it
>     to CPE, then CPE expects to get the same prefix back at least for that
>     period of time. That doesn't mean that you can assign prefixes at will
>     at any time you want. You are breaking the "contract" to your customer
>     (CPE). That's exactly the point where things break.
> 
>     If you set a lifetime of assigned prefix to 24 hours and then change it
>     - well, that's your decision, but you are not suppose to break the
>     timers that you set yourself.
> 
> 
> That's fine, except it doesn't really have anything to do with your 
> previous statement.  I can (and do) assign a single large prefix to each 
> BNG for the DHCPv6 server to assign dynamic PDs from, and still respect 
> the above assertions.

I said "where BNG/BRAS allocates prefixes to customers "at it's will"". 
I didn't say "after the lifetime has expired". But, nevertheless, I 
think you and I already had this conversation, didn't we?

You deeply understand what you are doing, you fixed your home-brew CPE 
(or vendors CPE for your use, it doesn't matter), implemented 
across-reboot PD memory and things work. Well done, congratulations!

But you are one operator on a big planet, where many other operators 
doesn't get nowhere near to your understanding of how to build that 
stuff and implement IPv6 in a proper way.

BTW, on a side note, hopefully you can come to SINOG this year and 
explain to our operators how you did exactly things that we are 
discussing here ;) (let's talk offline...)

> 
> We use 1 hour lease times.  If a CPE hasn't sent a Release, has the same 
> MAC address and DUID, it will continue to get the same PD after a 
> [hard|soft]-reboot.  In fact it can return anytime up to an arbitrary 
> value (7 days in my network), and still receive the same PD, if the 
> BNG/DHCPv6 server is configured to keep state for longer.
> 
> As just mentioned by Lee, we should be making sure that these protocols 
> are resilient to change, not dictating rigid network designs.

I'm with you. If we manage to fix CPE and host to be able to follow that 
dynamic topology changes - that would be a good thing to do anyway. I'm 
not at all encouraging operators to deploy dynamic PDs from many 
reasons, but if topology changes then CPE and hosts needs to be able to 
follow that change.

Cheers, Jan