If you have two connections to the same ISP from different sites, it's common to have an exit policy that is different at each site. Often, the policy at each site selects one interface on a border router at that site as the default exit interface. Exit traffic making it to any of the border routers for the site will use the default exit interface for the site. This policy might be called "use the closest default interface."
Choosing the closest exit often makes sence because your ISP's network is probably faster than yours. However, the best exit may not be the closest if for example:
The closest exit is chronically congested while you have excess capacity on your internal network and an alternate exit that is under used.
The closest exit is slower than your internal network and an alternate exit.
If you do have a fast private network between sites, and if some sites have Internet connections that're slower than the private network, such sites may send their exit traffic over the private network toward a site with a faster Internet connection. If the fastest Internet connection is at headquarters, then your exit policy may be called "send all exit traffic to HQ."
For example, consider the network shown in Figure 7-6. There is a fractional T1 to your ISP from a smaller site that's intended primarily as a backup for when a full T1 serving a main site goes down. There is also a private full T1 connecting your main site to the smaller site. Traffic originating on LANA can take the closest exit through Border RouterA and also be taking the fastest exit path. However, traffic originating on LANB may not be taking the fastest exit if it takes the closest exit via Border RouterB. It might make sence for some traffic originating on LANB and destined for the Internet to cross the T1 and take the exit through Border RouterA at the main site. This is especially true if your sites are near to each other since there's then likely to be little delay over your private T1 between them.
Choosing the right exit in this case is primarily the job of your IGP. (However, BGP should be providing crucial information that your IGP needs to make the choice.) There are two techniques that're commonly used:
Accept a default route via BGP from your ISP via each border router and inject this route into your IGP. See the section called Use Closest Interface with ISP Border Router Connectivity below for details.
Accept a BGP "backbone route advertisement" from your ISP via each border router and inject this route into your IGP. See the section called Use Closest Interface with ISP Core Router Connectivity below for details.
Which technique you choose will probably depend on the IGP you run, your ISP, and if some hosts on your network are listening to information from your IGP. These issues are discussed in following sections.
Copyright © 1999-2000 by Robert A. Van Valzah