I thought about why it might be best to have a single router
(e.g., DR) that floods each LSA and is also adjacent with all
other routers.  I think the following example shows that it is
best to have such a router (versus smart peering).

Consider a single-hop broadcast network with 3 routers having
RIDs 1, 2, and 3.  Assume 3 is adjacent to both 1 and 2.  If OSPF
(or draft-nsheth or OSPF-MDR) is used, router 3 will be the DR.
(Even though each router should have at least 2 adjacencies,
this uniconnected example will demonstrate my point.)

 3
/ \
1   2

Suppose router 1 floods an LSA, but none of the other routers
receive it.  Router 1 will retransmit the LSA to router 3
after RxmtInterval if either OSPF or draft-retana is used.
If OSPF is used, router 3 (the DR) will flood the LSA immediately
and router 2 will receive it.
However, if draft-retana (or RFC 5820) is used, router 3 will
not immediately forward the LSA (since it is not an MPR or
active OR), but will wait PushbackInterval plus jitter before
flooding the LSA.  This interval may be small (e.g., 2 seconds) but
as I mentioned in my last post, in large networks this mechanism
may result in several routers forwarding the same LSA unless
PushbackInterval is larger than stated in RFC 5820.

Suppose we fix this by having only one router/relay that
forwards the LSA when it is received from another router.
If, in the above example, that relay is router 2, then
it still would not work, since 1 is not adjacent with 2.
So the single relay must be router 3.  In general, the single
relay must be be adjacent with all other routers.
Therefore, one can conclude that it is best to have a special
router (e.g., DR) that is adjacent with all other routers.
Note that draft-retana does not have this property, and would have
to dispense with "smart peering" in order achieve this property.

Regards,
Richard

_______________________________________________
OSPF mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/ospf

Reply via email to