MPLS and Next-Generation Networks: Foundations for NGN and Enterprise Virtualization

One of the most common concerns of operators is the scalability of MPLS TE. Some of the common objections used are stated as follows:

Myth: MPLS TE uses RSVP and RSVP does not scale, so how can MPLS TE scale?

Reality: Actually, no per-flow reservation scales well in the network core. However for MPLS, the tunnels are created for all traffic between two nodes of a certain traffic type. For instance, all voice traffic flowing between two nodes flows through one tunnel. Moreover, RSVP used in MPLS TE is different from RSVP signaling used in integrated services (IntServ). In IntServ, RSVP reservations perform per-flow policing and queue reservation; in MPLS TE RSVP signaling is not associated with any queue reservation or policing, by design. Even the admission control is for an aggregate traffic flow between POPs and not per flow. In summary, RSVP for MPLS TE scales well with thousands of tunnels. Cisco has tested up to 2000 head ends and 15,000 midpoints on its core routers.

Myth: RSVP is stateful and is also a soft-state protocol. It requires too much state, and it needs to be refreshed periodically.

Reality: Yes, RSVP is a soft-state protocol. However, after the labels are distributed, the TE tunnel is in place until it is torn down. No per-flow policing is done, and the tunnel state is maintained with periodic refreshes. The protocol has been enhanced to add refresh reduction so that refresh messages can be bundled or aggregated together for efficient processing. This results in far fewer refresh messages. Moreover, with best design practices, this is not an issue and tunnel timers can be tuned to handle the pacing of refreshes and appropriate handling of messages.

Myth: Managing large number of tunnels is too difficult.

Reality: Yes, managing TE is an overhead but not to the degree it is assumed. TE and explicit routing provide some additional gains with respect to additional bandwidth inventory or link, node, and path protection against failures. Tools for management are available. Moreover, configuration is simplified with Autotunnel and Automesh, where the tunnels are set up automatically to the routers in the same domain.