[Ospf-wireless-design] Advantages of aligning adjacencies with relays

[Richard Ogier <[e-mail]>]

All,

As has been discussed, using Smart Peering results in adjacencies
that are not aligned with relays (MPRs or MDRs).
Of course, in a OSPF broadcast network, adjacencies are aligned
with the (Backup) DR, and perhaps there is a good reason for this.
So I have been thinking about possible advantages of keeping
this property of OSPF so that at least one endpoint of each
adjacency is a relay.  Here are some possible advantages.

1. Alignment helps take advantage of implicit ACKs.
If a flooding relay (MDR) has several adjacent neighbors,
then these neighbors can treat an LSA flooded by the relay
as an implicit ACK.  If Smart Peering is used with MPRs,
then in general an MPR has only a few adjacent neighbors (same
as any router), so only a few neighbors can treat a flooded
LSA as an implicit ACK.

2. Alignment helps take advantage of explicit multicast ACKs.
Every router must ACK every LSA explicitly or implicitly, so this
advantage refers to sending an explicit multicast ACK in response
to a duplicate LSA received as a unicast.  By requiring adjacencies
to be aligned with MDRs, a single multicast ACK sent by an MDR will
inform several (adjacent) neighbors that the LSA has been received,
thus suppressing a larger number of possible retransmissions than
if adjacencies were not aligned with MDRs.

3. Extendability to non-ackable LSAs (i.e., periodic flooding).
By having a fixed set of nodes responsible for flooding LSAs
(independently of the advertising router) and maintaining
synchronization with neighbors, it is easier to implement
options such as non-ackable LSAs (described in Appendix D of
the MDR draft), by having each MDR flood each LSA periodically
until a new instance has been received.  There are probably other
advantages to having a source-independent set of relays.

4. Here is an example that shows it can take longer for an LSA
to reach all routers if adjacencies are not alighed with relays.

1
|
|
2 --- 3
|     |
|     |
4 --- 5

In this example, nodes 2 and 3 are relays (MDRs or MPRs)
and all neighbor pairs are adjacent except for (2,4).
In particular, (4,5) is an adjacent pair even though
neither 4 nor 5 is a relay.
(I am making a general point so am not assuming that Backup
MDRs/MPRs are used.)
Suppose that node 1 floods an LSA and relay 2 forwards it.
Suppose nodes 3 and 4 do not hear 2's flood.
Then node 2 will retransmit the LSA to node 3 after 5 seconds,
but will NOT retransmit the LSA to node 4 (not adjacent).
Suppose node 3 receives the retransmitted LSA and relays it.
If node 5 does not receive it the first time, node 3 will
retransmit it after 5 seconds.  Now, since node 5 is not a relay
it does not immediately flood the LSA, but will retransmit it
to node 4 (adjacent) after 5 seconds.  This example shows that it can
take much longer for the LSA to reach node 4 than if adjacencies
were aligned with relays, so that the pair (2,4) is adjacent.

Richard



_______________________________________________
Ospf-wireless-design mailing list
[e-mail]
https://www1.ietf.org/mailman/listinfo/ospf-wireless-design