RIP Convergence

RIP Convergence
Convergence time is one of the problems associated with distance-vector routing protocols. This
section details the convergence process of the RIP protocol. We’ll use Figure 1.2 to help describe
the RIP convergence process.
The following list describes the RIP convergence events when a problem occurs. In Figure 1.2, the
WAN between Routers D and F goes down. This link was along the path from Routers A through
D, when delivering packets to the Ethernet segment off of Router F. Now, these four routers, in
particular Router D, must learn the path through Router E, but each of the four routers will notice
an additional hop to this network. Here’s what happens:
1. Router D poisons this directly connected route in its own routing table, removes it, and
sends a triggered update to Routers E and C. Any routes with Router D’s interface in that
downed link or Router F’s address on that link as a next hop will also be poisoned. This will
almost definitely include the Ethernet segment off of Router F.
FIGURE 1 . 2 Convergence
24 Chapter 1  Routing Principles
2. Router C was using this path to access Router F’s Ethernet, but Router E goes directly
through Router F. So the only effect on Router E is the poisoning and removal of the WAN
link from its routing table. Router C, however, poisons both routes, removing both of them
from its routing table. Router C sends a triggered update to Router B, which in turn sends
a triggered update to Router A, each router poisoning and removing the route locally.
Router F also sends a triggered update to Router E.
3. The triggered updates that Router E received from Router D and Router F prompt it to send
out its own triggered updates in each direction. These updates tell both of its neighbors that
all of the routes they used to have access to are still accessible through Router E. Router D
enters the new route for Router F’s Ethernet—through Router E—into its routing table.
4. Router D accepts this new route to the destination network, as it did not have it in holddown
because it was just purged; a route must be in the routing table in order to be in a
holddown state. Because route poisoning is in use, the same situation exists on the other
routers, as well. Very quickly, they will be ready to accept the new metric.
5. Router D advertises the new metric to Router C, and even if Router C had not poisoned and
removed the route from its routing table, but instead had this route in holddown, it would
accept the new metric, because it is from the source of the original advertisement—Router D.
6. The same effect trickles down to Routers C, B, and A in that triggered updates cause route
poisoning and subsequent triggered updates. So none of them have the failed route in their
routing table, nor do they have any of the routes that were accessible through the failed
link, including Router F’s Ethernet network, which allows their routing table entries to be
updated almost immediately by the less desirable metrics. In other words, even if the advertiser
of the original route was not advertising a worse metric, which itself is grounds for
believing the worse metric, the fact that poisoning removed the routes from their routing
tables makes these new updates look like routes they never heard of, causing them to be
learned without incident.
Without triggered updates and route poisoning, the time required for Router A to converge
would be the detection time, plus the holddown time, two update times, and another update
time. The complete convergence to Router A could be over 240 seconds. With these mechanisms
working for you, however, convergence will occur in an internetwork of this size in about
15 seconds. It will benefit you, though, to understand the logic behind the scenario that results
in the convergence time of roughly 240 seconds.
If the network comes back up, both Routers D and F would immediately broadcast
RIP requests for the neighbor’s entire routing table, toward each other
across the link between them in order to try to speed up convergence without
taking a chance on waiting for the other device to start advertising across the
link on its own. Instead of the normal broadcast or multicast, the response is a
unicast back to the requesting interface. The industrious reader could model
this scenario in a lab environment, if available. You could intentionally administratively
shut down an interface, while watching the results of the debug ip
rip command, and then bring the interface back up, allowing all that has been
presented here to be observed.