Advertising both the Summary route, and a the component routes can provide for optimal routing and fast convergence. The component routes provide the optimal routing (longest match rule), and the summary route is already there in the IP routing table to provide a path should a component route disappear.
In general, you should always have a link between two routers configured with the same Aggregate/Summary that is not configured for aggregation. This prevents potential black holes caused by the discard (Null0) route.
Optimal path selection is important for delivering the right service level to an application. For example, the high speed lower delay links may not be the best path for jitter sensitive traffic (voice) if those links are more heavily loaded than a medium speed, higher delay, but lightly loaded path. The voice traffic bandwidth requirements are not high, and medium amounts of delay are acceptable. The jitter introduced by heavily loaded links can be more detrimental to voice traffic than anything else.
The more routes a router has will increase convergence times because of larger topology databases to process for best path algorithms. The more parallel links for the sake of redundancy also increases convergence times to a point of diminishing returns. Keeping redundant paths to (2) parallel links and route summarization/aggregation where it makes sense is good practice for resilient networks that can converge quickly around failures.
- Link-state protocols hide topology information at the edge of a flooding domain.
- Distance vector protocols hide topology information at every hop.
- Path vector protocols hide topology through filtering and aggregation.
Summarizing may not always hide topology changes. If the summary includes components of different metrics the router may choose the highest or lowest metric of the components to use as the metric for the summary. If the component subnet from which the metric was drawn from fail, the Summary will need to be recreated with a new metric value, causing the upstream routers to notice the change.
Create a component subnet from a loopback interface the point of summary, and have the summary metric anchored to the loopback interface.
downside: the router might loose all reachability to subnets within the summary but still advertise the summary because the loopback interface is still within the summary.
Set the summary metric manually via configuration.
Its better to use static routes at the edge of your network and redistribute the statics into your IGP, rather than redistributing from IGP to IGP
Frame Relay fast convergence
Asynchronous LMI is when the Frame Relay switch tells the attached router about circuit failures before waiting for the next LMI polling interval.
If the frame relay DLCI is provisioned to a point-to-point subinterface, the LMI message indicating circuit failure will trigger an interface failure and notify the IP routing protocol.
Failed frame relay circuits (DLCI) associated to a multipoint interface will not trigger an interface down event, and therefore notifying the IP protocol of the failure will need to wait until the protocol HOLD/DEAD timer expires.
end-to-end LMI is another option along with Asynchronous LMI.