Re: [v6ops] SLAAC host moving between networks

[Mark Smith <[e-mail]>]

On Fri, 9 Nov 2018 at 20:31, Richard Patterson <[e-mail]> wrote:
>
> On Fri, 9 Nov 2018 at 02:42, Mark Smith <[e-mail]> wrote:
> > BRASes and CPEs both send keepalives to detect this event if it is
> > disruptive to packet flow.
>
> When using PPP[oE], sure, but what about IPoE?

So I don't have first hand experience with IPoE, however I have to
assume that since it has been used for HFC networks universally for in
the order of 15+ years, and a broadband ADSL network I know of for the
past 10 or more years, that is fundamentally a solved problem.
Otherwise it would be well known that there are hanging
sessions/RADIUS accounting problems with HFC/IPoE networks, and I
expect it would have fallen out of favour as an access method by now.
There's been plenty of time for that to happen.

The bit of research I've done in the past (about 10 years ago now),
and verified now is that they're using things like ARP ping and ICMP
echo request/response in addition to DHCP to create and maintain
session state.

For example, the copyright date on this presentation is 2007, and that
describes how IPoE new sessions are established and how session
failure is detected via a number of methods (DHCP lease expiry, ARP
ping failure, ICMP ping failure).

http://www.swinog.ch/wp-content/uploads/2018/07/The_Evolution_from_PPPoE_to_IPoE_sessions.pdf

And here is a description of how another vendor is doing it (the same way)

https://www.juniper.net/documentation/en_US/junos/topics/concept/subscriber-management-dhcp-liveness-detection.html

> A problem we're trying to solve with draft-patterson-intarea-ipoe-health
>

I think the only and a major criticism of the above methods is that
they're using BRAS control plane resources for this liveness testing,
and that can become considerable with 1000s of subscriber sessions on
a BRAS.

Having CPEs use forwarding plane verification i.e. sending traffic
that is looped through the forwarding plane of the BRAS, back to the
CPE, rather than CPEs using keepalives methods that are processed by
the control plane of the BRAS (including PPP LCP Echo Request/Reply in
a PPPoE scenario), would help remove a useful amount of load from BRAS
control planes.

Regards,
Mark.


> -Richard