Extract from Wendell Odom's "Official ROUTE Certification Guide" with some omissions:
"The final concept to consider when summarizing routes is that the packets may take a longer path than if summarization
is not used.Routers that learn a summary route do not know about the details of the subnets inside the summary. Instead,
the routers pick the lowest metric summary route for a prefix. Later, when these packets arrive at routers that do know all
the subnets inside the summary, those routers can then use the best routebe it a short route or long route.
When EIGRP advertises a summary route, the advertising router considers the summary route to be up
and working unless all subordinate routes fail.
For example, next Figure shows the less efficient routing of packets to B1 through WAN1, instead of passing directly to
WAN2. When WAN1’s 768 Kbps CIR PVC to Router B1 fails, WAN1 does not change its route advertisement for its summary route. So, Core1 and Core2 use WAN1 as their next-hop router for their routes to
Suboptimal Forwarding Path when Primary PVC Fails"
How could the routing described in the previous figure ever happen, considering the situation as it has been described
in this section (Route Summarization) where WAN1 and WAN2 are in fact connected to many other branch routers
accessing and networks, and WAN1 is the next-hop router for Core1 and Core2 ?
1) No stub routers
If no branch router is configured as a stub router, then all these branch routers can route traffic between
neighbors WAN1 and WAN2 and in this case, when the WAN1 çè B1 link goes down, WAN1 has alternate
routes through all other branch routers to which it is connected (B2, B3…) and not through WAN2.
2) All branch routers are designed as stub (optimizing the DUAL algorithm)
a. By default, the AD of summary routes is 5, which means WAN1 cannot have an alternate route for B1
through WAN2 (AD 90), assuming there are some routes within the summary route still UP,
This means that if only WAN1 B1 link goes down (and other links from WAN1 within the same summary
route remain UP), B1 becomes unreachable from the rest of the network, whereas there are physical
alternate paths !
b. If the AD of the summary route on WAN1 is changed to 90 to give a chance to B1 to become accessible
again, there would be a risk to introduce some routing loops between WAN1 and WAN2, assuming that
WAN2 has the same summary route.
The GNS3 Lab Suboptimal Forwarding Issue with Summarization that I designed can be used to check
this point (it covers other features as well which are irrelevant to this situation). It uses 7206 and 3725
router images but you can easily change this to suit your needs.