In message <[email protected]>
Pierre Francois writes:
> Areas in a link-state IGP turn it into a mix of distance-vector
> and link-state protocols. This makes a simple link event in one
> area look like a SRLG event in the others, as the rerouting routers
> inside an area will need to receive all the updated summary-LSAs of
> their Area Border Routers to be allowed to perform
> their final FIB Updates. Moreover, a single link event in an area A can
> lead to transient FIB updates in an area B to finally return to the same
> fowarding states in B as before the event. It's impossible to have those
> transient unecessary FIB updates/loops without area.
> Moreover, it puts some restrictions on the possible paths between a
> source and a destination, according to the network topology, as a S-D
> path cannot leave and re-enter a given area although this path could be
> the optimal one.
> Now that we are convinced that the SPF recomputation time is not
> the critical time-component influencing the convergence of an IGP, why
> don't we reconsider the usage of mutiple areas inside a network where
> the convergence time is considered important ?
> However, OSPF convergence with Stub areas shoudln't be a problem, as
> we only have to ensure that packets are consistently forwarded outside
> the stub and that forwarding inside area 0 is also consistent to reach
> the area Border Routers for the destinations that are in stubs.
> Transit areas make the problem similar to the convergence of BGP between
> ASes, except that routers don't have complete paths but only distance
> information. Aren't the benefits of IGP areas historical ?
Note that for MPLS (RSVP/TE) FRR if the advertised cost is the cost of
the tunnel within a given area, then the effect of a link failure in
one area is confined to that area alone. No ASE or summary routes
need to be changed because in effect there is no cost change within
the area. To acheive this tunnels are configured to be treated as
links by the IGP and the cost of the tunnel is configured rather than
dependent on the cost of the path taken at any given time.
Of course since avoiding the use of MPLS is a religious tenant of this
WG, please disregard this point and carry on with the discussion. :-)
Rtgwg mailing list