6.4 Scaling Dial-on-Demand Routing
6.4.6 Routing Strategy
The development of the routing strategy for this network is based on the following two requirements:
  • When a particular remote site is not dialed in to the central site, that remote site must be reachable through any central-site access router by means of a static route configured in each central-site access router.
  • When a particular remote site router is logged in to a central-site access router, that remote site must be reachable through that central-site access router by means of the dynamic route that has been established for that connection and propagated to the backbone.

To meet these requirements, the central-site access routes advertise the major network route of the remote sites to the Cisco 4500. All routes to the remote sites are equal-cost through all of the central-site access routers. Each central-site access router is configured to have a static route to each remote site. To allow the Cisco 4500 to use all of the central-site access routers for connecting to the remote sites, the no ip route-cache interface configuration command is configured on Ethernet interface 0 of the Cisco 4500, disabling fast switching of IP to the subnet shared with the central-site access routers. This causes the Cisco 4500 to alternate between the three access routers when initiating outbound calls. This strategy increases network reliability for those cases when one of the access routers goes down.

When a remote router logs in, it announces its IP address and sends a RIP flash. The RIP flash causes a dynamic route to the remote site to be installed immediately in the routing table of the central-site access router. The dynamic route overrides the static route for the duration of the connection.

Next, the central-site access router redistributes the RIP route into Open Shortest Path First (OSPF) and sends the route to all of its OSPF neighbors, including the Cisco 4500, which installs it in its routing table. The Cisco 4500 now has a major network route to all of the remote sites, plus a dynamic route to the specific remote site that has logged in. If a central-site host needs to communicate with a particular remote site that is currently logged in, it does so through the dynamic route.

When the remote site logs out, the dynamic route must be removed from the Cisco 4500, and the static route to the remote site must be restored on the central-site access router into which the remote router logged in.

If a central-site host requires communication with a remote site that is not logged in, it will use the major network route defined in the Cisco 4500. A central-site access router, selected in round-robin fashion, is used to initiate the call to the remote site via the static route that is defined for it in the configuration for the selected access router. As in the case of a remote site that calls the central site, once the connection is made, the remote-site router sends a RIP flash that causes a dynamic route to the remote site to be installed immediately in the routing table of the central-site access router. This dynamic route is redistributed into OSPF and is installed in the routing table of the Cisco 4500. Figures - use a state diagram to summarize the routing strategy.

The following convergence issues pertain to the state diagram shown in Figures - :

  • During the time between State 2 and State 3, a host at the central site might initiate a call to the remote site. Until State 3, at which time the routes converge on the Cisco 4500, any central-site access router that dials the remote site will fail with a busy signal. In practice, only one call fails: by the time a second connection attempt is made, the routes will have converged in State 3, the dynamic route will be available for use, and there will be no need to make another call.

If, before convergence occurs, the Cisco 4500 directs a call through CENTRAL-1 to Router 2, the call will fail and must be retried. IP fast switching is turned off on the Cisco 4500, so the Cisco 4500 (which is using equal-cost paths to each of the central-site access routers) will send the next packet through CENTRAL-2 or CENTRAL-3 (which still have a static route for Router 2) and the call will go through.

Note: When developing the routing strategy for this network, the designers considered the use of snapshot routing, which reduces connection cost by limiting the exchange of routing protocol updates. For snapshot routing to work, each remote site must connect to the same access router every time it dials into the central site. In this design, the remote routers connect to the central-site access routers through a hunt group, so there is no way to control to which central-site access router a remote router will connect for any particular connection. Therefore, snapshot routing cannot be used for this design.