Frame Relay is unlikely to be cost-effective when only two sites are interconnected with a point-to-point connection. Frame Relay is more cost-effective where multiple sites must be interconnected.

WANs are often interconnected as a star topology. A central site hosts the primary services and is connected to each of the remote sites needing access to the services. In this hub and spoke topology the location of the hub is chosen to give the lowest leased line cost. When implementing a star topology with Frame Relay, each remote site has an access link to the frame relay cloud with a single VC. The hub has an access link with multiple VCs, one for each remote site. Because Frame Relay tariffs are not distance related, the hub does not need to be in the geographical center of the network.

A full mesh topology is chosen when services to be accessed are geographically dispersed and highly reliable access to them is required. With full mesh, every site is connected to every other site. Unlike with leased line interconnections, this can be achieved in Frame Relay without additional hardware. It is necessary to configure additional VCs on the existing links to upgrade from star to full mesh topology. Multiple VCs on an access link will generally make better use of Frame Relay than single VCs. This is because they take advantage of the built-in statistical multiplexing. 

For large networks, full mesh topology is seldom affordable. This is because the number of links required for a full mesh topology grows at almost the square of the number of sites. While there is no equipment issue for Frame Relay, there is a limit of less than 1000 VCs per link. In practice, the limit will be less than that, and larger networks will generally be partial mesh topology. With partial mesh, there are more interconnections than required for a star arrangement, but not as many as for a full mesh. The actual pattern is very dependant on the data flow requirements.

In any Frame Relay topology, when a single interface is used to interconnect multiple sites, there may be reachability issues. This is due to the nonbroadcast multiaccess (NBMA) nature of Frame Relay. Split horizon is a technique used by routing protocols to prevent routing loops. Split horizon does not allow routing updates to be sent out the same interface that was the source of the route information. This can cause problems with routing updates in a Frame Relay environment where multiple PVCs are on a single physical interface.

Whatever the underlying topology of the physical network, a mapping is needed in each FRAD or router between a data link layer Frame Relay address and a network layer address, such as an IP address. Essentially, the router needs to know what networks are reachable beyond a particular interface. The same problem exists if an ordinary leased line is connected to an interface. The difference is that the remote end of a leased line is connected directly to a single router. Frames from the DTE travel down a leased line as far as a network switch, where they may fan out to as many as 1000 routers. The DLCI for each VC must be associated with the network address of its remote router. This information can be configured manually by using map commands. It can also be configured automatically, taking LMI status information and sending a Reverse Address Resolution Protocol (RARP) message on each VC identified. This process is described in more detail in a separate section.

Web Links Configuring Dynamic and Static Mapping for Multipoint Subinterfaces http://www.cisco.com/warp/public/ 125/ 17.html#17-A