Summary

In this chapter, we've discussed the specifics of running MPLS across ATM networks. The MPLS architecture allows MPLS to be deployed in ATM networks with no hardware upgrades to the ATM switches.

ATM switches do need new control software in the control processors that support MPLS signaling. Some switches cannot support the increased demands, resulting in the need for an external controller (Label Switch Controller) that provides MPLS support for such a switch.

The MPLS forwarding and label allocation procedures were slightly modified to support the ATM environment:

• Cell-based label switching is performed purely based on VPI/VCI values in ATM cell headers to support the existing ATM infrastructure. The top-of-stack MPLS label is thus encoded in the ATM cell header.

• Even though the top-of-stack label is moved into the ATM cell header, the MPLS stack in the labeled packet is still intact because it is needed to support additional MPLS functionality such as MPLS experimental bits or the TTL field. The label in the top entry of the MPLS label header is set to 0 because it is not used across an ATM network.

• Label distribution in an ATM network is based on downstream-on-demand procedures to minimize VC usage across LC-ATM interfaces.

• Traditional ATM switches must request a label from the downstream LSR before they can allocate a label to an upstream LSR and establish inbound-to-outbound VPI/VCI mapping in the ATM switching matrix. A new label must be requested from the downstream LSR for each upstream request to prevent cell interleave problems.

• Advanced ATM switches support VC merge, additional cell buffering that prevents cell interleave problems. These switches can use the same downstream label for all upstream neighbors, resulting in significant savings of VCs used across LC-ATM interfaces.

The downstream-on-demand label distribution in ATM networks also affects the convergence time of ATM-based MPLS networks. The overall convergence time usually increases because new labels must be requested and allocated following the convergence of an IP routing protocol.