What Is MPLS TE?
In recent years, we’ve seen many networks migrate from multitechnology designs where the network core was implemented with Frame Relay or ATM and the network edge was implemented with routers to pure-router designs first deployed by Internet service providers. As the networks were being migrated, network managers quickly discovered a limitation of IP routing that was well known to its early adopters – after two traffic flows for the same destination are merged, it’s impossible to split them and route them over different paths.
This limitation can be easily demonstrated in a simple network, as shown in Figure 1. The traffic from router A to router C and the traffic from router B to router C are merged in router D and flow over the same link (D to E), potentially resulting in an overloaded link while the other links in the network might be completely unused.
)
Figure 1: Suboptimal Traffic Flow in a Simple Network
It’s almost embarrassing to notice that the earlier networks based on Frame Relay or ATM in their core did not have this limitation because they used connection-oriented end-to-end circuits that could be easily routed across any desired path in the network. The same network topology, with core routers being replaced by Frame Relay switches, solves the link overload problem (as shown in Figure 2) by placing the virtual circuits A to C and B to C over different core links.
)
Figure 2: Optimal IP Flow in ATM-Based Network Core
Some readers might argue that you can always reach a close-to-perfect IP-only network topology if you know your traffic patterns and if you can install the core links on an as-needed basis. Unfortunately, most of us are not that lucky – we have to live with what we have. Clearly, we need the capability (called Traffic Engineering) that we had in Frame Relay or ATM in the IP-only world.
Implementation of TE usually requires the ability to set up end-to-end virtual paths across the network – something that was never available in connectionless IP networks. With the introduction of MPLS, which effectively provides virtual circuits (called Label Switched Paths [LSP]) across IP networks, it was only a question of time when MPLS would be used to implement proper Traffic Engineering in IP-only networks. MPLS TE follows the same principles as ATM Traffic Engineering. The only difference is the choice of protocols used to implement them:
The network manager defines traffic trunks (implemented as tunnel interfaces) on the routers at the network edge. (ATM analogy: SVC setup parameters are defined on ATM edge routers.)
Link-state routing protocols (OSPF or IS-IS) make the network topology and the resource availability known to all nodes in the network. (ATM analogy: PNNI protocol.)
Modified Resource Reservation Protocol (RSVP) is used as a signaling protocol (ATM analogy: UNI signaling) to set up LSPs (ATM analogy: ATM virtual circuit) across the network.
The similarity between MPLS TE and ATM might prompt another question: If these technologies are so similar, why should I migrate to a router-only network? The answer is simple: You reduce the number of platforms you have to manage, the number of network management systems, and you reduce the scope of knowledge needed by your engineers. This results in lower training and operational costs.
The usability of MPLS TE extends beyond the router-only networks. Even users that have routers connected with Frame Relay or ATM virtual circuits can use MPLS TE to shift the traffic load onto less-utilized virtual circuits, as shown in Figure 3. The network shown in this figure has three core nodes (routers) connected with Frame Relay DLCIs of various speeds. MPLS TE can treat each DLCI as a leased line with user-specified bandwidth (usually equivalent to Committed Information Rate [CIR]) and balance traffic trunks based on the available bandwidth on each core link.
)
Figure 3: MPLS TE Deployed Across a Third-Party Frame Relay Network