Re: Persistent loops when mixing rtgwg-enterprise-pa-multihoming and rtgwg-dst-src-routing

[David Lamparter <[e-mail]>]

On Wed, Jul 26, 2017 at 12:16:17AM +0000, Acee Lindem (acee) wrote:
> I must admit that I had always thought that the source-routing paradigm in
> draft-troan-homenet-sadr-01.txt was backward with the destination address
> Longest Prefix Match (LPM) being done prior to the source address lookup.
> Rather I think if were going to standardize in the RTG WG, it should be
> the FIB organization described in section 3 of
> draft-ietf-rtgwg-enterprise-pa-multihoming-01.txt. Note that doing the
> source address lookup first  maps directly to the PA multi-homing
> use-case.

I see both a misunderstanding and an internal contradiction here.

The misunderstanding is that enterprise-pa-multihoming is describing
source-first RIB or protocol behaviour.  It isn't - it's describing a
destination-first RIB and a source-first FIB translation.  Conversely,
rtgwg-dst-src-routing isn't specifying a dst-first FIB, see below.

This is also where this contradicts itself;  the PA multi-homing
use-case does *not* map to source-first.  If it did, you wouldn't need
the route duplication.  In fact, the PA multihoming case -- just like
homenet -- needs destination-first to get local connectivity up and
running.

Also, and very importantly, I really don't want to go around telling
people how to run their FIBs.  It doesn't matter, as long as they're
forwarding packets in a manner confirming to the spec.  This is also why
I've moved A.1 and A.2 into the appendix on the -05 rev of the draft;  I
felt these suggestions on how to efficiently operate this on a
source-first (A.2) or destination-first (A.1) FIB are most appropriate
in an appendix.

On Wed, Jul 26, 2017 at 08:52:53PM +0000, Acee Lindem (acee) wrote:
> I believe the tables could be similarly collapsed giving source address
> higher precedence than destination address. Do you disagree?

They can be, this is described in appendix A.2. of dst-src-routing-05.
This does result in more duplicated routes than the other way around,
but it's a perfectly viable implementation choice.

That said, it makes no sense to use this in routing protocol signaling,
or even the RIB.  This is a FIB implementation detail.  For protocol and
RIB operation, destination-first-with-fallthrough provides the semantics
that make the most sense to describe a network and get it up and
running -- even if all of the routers have source-first FIBs and perform
this translation step on installing a route.

Cheers,


-David