Re: Generic anycast addresses...

On Fri, 31 May 2019 at 10:15, Fred Baker <fredbaker.ietf@gmail.com> wrote:
>
>
>
> > On May 30, 2019, at 2:59 PM, Ted Lemon <mellon@fugue.com> wrote:
> >
> > Do ULAs have a non-fuzzy scope boundary?
>
> The recommendation of RFC 4193, if I understand it, is that advertisements of ULA prefixes in EBGP are "not announced", and if they happen to be announced by accident, are refused. https://tools.ietf.org/html/rfc4193#section-4.1.
>

>From memory, I suggested that at the time, and the motivation was to
try to avoid by default ULA private address space route leaks we've
seen with the RFC 1918 address space, more recently with 100.64/10,
and other non-global address spaces.

The idea was to facilitate trading ULA routes through more specifics
if e.g. two ULA domains were to be joined together after a company
merger, but to not have it occur automatically upon bringing up eBGP.

BCP38 would also be another cause of ULA packet drops at customer/ISP
network boundaries.

Regards,
Mark.

> That is in no sense a default. My primary expertise is with my former employer's equipment, but I believe this to be generally true: EBGP only advertises what it is told to, and refuses only what it is told to. Hence, I would guess that the typical configuration will not advertise a ULA in anything resembling a normal case, but will not refuse one that is advertised to it.
>
> But the intended scope-boundary is "EBGP".
> --------------------------------------------------------------------------------
> Victorious warriors win first and then go to war,
> Defeated warriors go to war first and then seek to win.
>      Sun Tzu
>
> --------------------------------------------------------------------
> IETF IPv6 working group mailing list
> ipv6@ietf.org
> Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
> --------------------------------------------------------------------