Layer 3 flows over Frame Relay:

* Choices for Layer 3 addresses on Frame Relay interfaces.

* Broadcast handling.

In particular, the Frame Relay implementation in Cisco defines three different options for assigning subnets and IP addresses on Frame Relay interfaces:

* One subnet containing all Frame Relay DTEs.

* One subnet per VC.

* A hybrid of the first two options.

Frame Relay Layer 3 Addressing: One Subnet Containing All Frame Relay DTEs

In a full mesh, each router has a VC to every other router, meaning that each router can send frames directly to every other router.  This more closely resembles how a LAN works.  So a single subnet can be used for all the routers’ Frame Relay interfaces, as configured on the routers’ serial interfaces.

The single-subnet alternative is straightforward, and it conserves your IP address space.  It also looks like what you are used to with LANs, which makes it easier to conceptualize.

Frame Relay Layer 3 Addressing: One Subnet Per VC

The second IP addressing alternative, having a single subnet for each VC, works better with a partially meshed Frame Relay network.

The single-subnet-per-VC alternative matches the logic behind a set of point-to-point links.  Using multiple subnets instead of one larger subnet wastes some IP addresses, but it overcomes some issues with distance vector routing protocols.

Frame Relay Layer 3 Addressing: Hybrid Approach

The third alternative for Layer 3 addressing is a hybrid of the first two alternatives.  Two options exist for Layer 3 addressing in this case.  The first is to treat each VC as a separate Layer 3 group.

Point-to-point subinterfaces are used when a single VC is considered to be all that is in the group.  Multipoint subinterfaces are used when more than two routers are considered to be in the same group.

Multipoint subinterfaces logically terminate more than one VC.  The name “multipoint” implies the function, because more than one remote site can be reached via a VC associated with a multipoint subinterface.

Most of the time, point-to-point subinterfaces are used, with a single subnet per PVC.