From: Chuck Church (cchurch@optonline.net)
Date: Sat Dec 07 2002 - 22:19:49 GMT-3
I don't believe that subinterfaces and point-to-point OSPF would have this
problem.
Chuck Church
CCIE #8776, MCNE, MCSE
----- Original Message -----
From: "OhioHondo" <ohiohondo@columbus.rr.com>
To: <ccielab@groupstudy.com>
Sent: Saturday, December 07, 2002 7:24 PM
Subject: Observation on OSPF in a Full Mesh -- Comments Requested
> Everybody
>
> I always assumed that the full mesh environment would "self heal" with the
> loss of a single pvc in the mesh. I found this not to be the case. I
tested
> two instances using both OSPF network types "broadcast" and
"non-broadcast".
> If the findings below are true the advantages of having a full mesh only
> apply to when an entire sync interface goes down?
>
> 1 - A pvc between two non-DR/BDR routers fail.
> When this occurs traffic between the two non-DR/BDR routers is
non-existant
> even though the IP routing tables of the routers indicate that there is a
> path. The reason seems to be that the non-DR/BDR routers receive their
view
> of the network from LSA's provided by the DR. The DR shows both of the
> non-DR/BDR routers as active in a network, LSA type 2 update. The
non-DR/BDR
> routers build IP routing table entries showing the next-hop to one another
> as direct to the others sync interface on the frame relay. That pvc is
down
> so no traffic flows between them.
>
> 2 - The pvc between the DR and the BDR fails.
> When this happens the BDR may keeps its' neighbor relationship with the
> non-DR/BDR routers and continue to receive updates from them OR if one of
> the non-DR/BDR routers becomes BDR then the old-BDR router is totally
> divorced from the network. In any case the BDR loses all connectivity with
> the DR and therefor loses DR routes in its' routing table and doesn't
> receive updates from the DR. Bad, bad, bad.
>
>
> If an entire sync link is lost instead of a single pvc, the remaining
> routers are still in a full mesh. They act accordingly, electing a new DR
> and BDR if necessary. They do not reflect IP routes as accessible when
they
> are not -- as in the first instance above.
> .
.
This archive was generated by hypermail 2.1.4 : Fri Jan 17 2003 - 17:21:40 GMT-3