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

[Richard Ogier <[e-mail]>]

Hi Lili,

Your proposed change assumes that the non-DR router knows when the
DR has finished sending LS request packets.  I am not sure this can
be assumed, since a router doesn't indicate when it is finished
sending LS request packets (and might not send any).
I think the only indication is when the DR declares the adjacency
to be Full when the LoadingDone event occurs (and adds the neighbor
to its LSA).

Even if the non-DR router determines via Database Description
packets that the DR has out-of-date LSAs, the DR could receive
those LSAs from another router.

It seems easier to just require that the DR's LSA include a link to
the router itself (router i in my example), or to the other non-DR
router j (which is what I used in my draft), before router i includes
a link to j in its router-LSA.  I prefer the latter because then
router i advertises a link to j only if router i is Full with the DR
and the DR is Full with router j.  Thus, by transitivity router i is
equivalent to being Full with router j (i.e., router i is as
up-to-date as router j).

Richard


Lili Wang wrote:

>Hi Richard,
>
>   Sorry for the late reply!
>
>Yes, this is a problem with the current spec.  We incorrectly convinced ourselves that such a check is not required.
>
>Here's the proposed change.
>
>We make sure the non-DR router have a synchronized database with the DR before advertising the Type 1 link to other non-DR router.
>We declare the DR as a fully synchronized adjacency after having all the DR's LSA request satisfied (i.e., Our LSA update for the LSA requests are acknowledged by the DR.) in addition to having a full adjacency with the DR. 
>
>The rule then will look like the following:  
>
>o  If a router is not the DR and has a fully synchronized adjacency to the DR, it MUST add a Type 1 link corresponding to each neighbor that is in
>state 2-Way or higher.
>
>
>Tanks,
>
>Lili
>-----Original Message-----
>From: [e-mail] [mailto:[e-mail]] On Behalf Of Richard Ogier
>Sent: Tuesday, May 17, 2011 12:05 PM
>To: [e-mail]
>Subject: Re: [OSPF] OSPF Hybrid Broadcast and P2MP Interface Type
>
>Section 3.6 of draft-nsheth-ospf-hybrid-bcast-and-p2mp-01
>has the following rule:
>
>o  If a router is not the DR and has a full adjacency to the DR, it
>   MUST add a Type 1 link corresponding to each neighbor that is in
>   state 2-Way or higher.
>
>Suppose i and j are non-DR routers.  It seems to me that i should
>also require that the DR's router-LSA include a link to j before
>i adds a Type 1 link corresponding to j, in order to ensure that
>i and j are fully synchronized before either uses the other as
>a next hop.  Is there a reason why this condition was omitted?
>
>To explain further, the SPT calculation (Section 16.1 of RFC 2328)
>requires that router j advertise a link back to router i before i
>can use j as a next hop (and vice versa). Thus, routers i and j can use
>each other as a next hop if they both advertise a link to each other.
>
>Therefore, the above rule from draft-nsheth only ensures that
>routers i and j are fully adjacent with the DR before either can
>use the other as a next hop.  As a result, the DR might not be fully
>adjacent with router i or j, and thus i and j may not be fully
>synchronized.  Note that full adjacency with a neighbor does not
>imply that the link state databases are synchronized (see footnote 23
>in RFC 2328).  It only means that the router is at least as
>up-to-date as the neighbor, since it only means that all Link
>State Requests have been satisfied.
>
>Please correct me if I am mistaken.
>
>Richard
>_______________________________________________
>OSPF mailing list
>[e-mail]
>https://www.ietf.org/mailman/listinfo/ospf
>
>  
>