Re: [OSPF] OSPF Hybrid Broadcast and P2MP Interface Type

[Richard Ogier <[e-mail]>]

I decided to read the posts regarding this draft (hybrid-bcast-p2mp),
since it is related to OSPF-MANET, and have a few comments.

It was mentioned that RFC 5820 (OSPF-MANET with smart peering) solves
the problem of this draft, assuming that smart peering and unsynchronized
adjacencies are used. (E.g., Sheth's post on 2/10/2011.)

I also want to point out that RFC 5614 (OSPF-MANET-MDR or OSPF-MDR),
which uses CDS flooding via MDRs (MANET Designated Routers), not only
solves the problem of this draft, but provides almost the same solution
as this draft when OSPF-MDR (with full-topology LSAs and biconnected
adjacencies) is applied to single-hop broadcast networks.

The hybrid-bcast-p2mp draft is nice because it requires only minimal
modifications to OSPF, keeping the DR and BDR and the same adjacencies.
The reason that OSPF-MDR applied to single-hop broadcast networks
provides almost the same solution as this draft is because it
generalizes the concept of DRs and BDRs to MANETs, and requires
only 2 adjacencies for a non-DR/BDR (or non-MDR/BMDR) in a
single-hop broadcast network.  More on this (near) equivalence below.

In contrast, the solution of draft-retana-ospf-manet, which applies
RFC 5820 with smart peering to single-hop broadcast networks, is quite
different, resulting in a more random selection of adjacencies,
which are not common to any single node such as a DR.
Thus, unlike the other two solutions, draft-retana-ospf-manet is
conceptually quite different from OSPF broadcast, since it does not
use the concept of the DR and BDR.

One of the goals in the design of OSPF-MDR was to minimize changes
to OSPF.  Thus, instead of throwing out the DR and BDR, they were
generalized to multi-hop networks, and were kept essentially the
same in the special case of a single-hop network, for the purpose
of flooding and forming adjacencies.  More information on OSPF-MDR
can be found at www.manet-routing.org.

Now more on the (near) equivalence between hybrid-bcast-p2mp and
OSPF-MDR when applied to a single-hop broadcast network.
In this case, the MDR selection algorithm is very similar to the
DR election algorithm of OSPF, and both select a single DR/MDR.
(There are a few minor differences that are not important to
this discussion.)  Also, if AdjConnectivity = 2, each node forms
an adjacency with the MDR and BMDR.

The origination of router LSAs is also nearly equivalent.
The rules in Section 3.6 of hybrid-bcast-p2mp, which specifies
which Type 1 links a non-DR includes in its router LSA,
are similar to the corresponding rules in RFC 5614 when
LSAFullness = 4 (full-topology LSAs).  In that case, a router
includes all bidirectional (2-Way or higher) neighbors that are
"routable", where a neighbor is routable if SPF has calculated
a route to the neighbor.  Note that SPF will calculate such a route
to the neighbor as long as both the router itself and the neighbor
are fully adjacent to the MDR.

This is slightly different from Section 3.6 of hybrid-bcast-p2mp, which
only requires that the router itself be fully adjacent to the DR (thus
relying on the requirement of SPF that the neighbor must include a
link back to the router in its LSA).  But both solutions achieve the
same goal, in a similar way, using a single DR/MDR that becomes
adjacent with all neighbors.

Richard