12.2 Traffic-Engineering Solutions

Although the concept has been around for some time, traffic engineering in IP is still in its early stages because Internet usage only began to truly push the edge of the envelope in the middle to late 1990s. This strain on resources forced the rapid development of feasible solutions. The three most prevalent of these in the networking world are Routed IP, switched transport, and MPLS. These solutions are described in the following sections.

12.2.1 Routed IP

IGP routing protocol solutions are used to route IP, and as they have been extended, they have been used to manipulate how traffic flows through the network. By tweaking metrics on specific links, it is possible to cause certain links to look more desirable than others. One major problem with metric manipulation is the possibility of unwanted side effects. For example, modifying link metrics in one part of a network may cause traffic to be routed suboptimally through other parts , as was demonstrated in Figures 12-1 and 12-2.

By itself, IGP route manipulation works well in small networks of limited complexity, where the effects of metric tweaking would not be as widely felt and would definitely be more manageable and predictable than in a large-scale network. While IGP metric tweaking is still possible and indeed currently implemented in larger-scale networks, it is not the most accurate or predictable approach available.

12.2.2 Switched Transport

Switched-transport networks, also known as overlay networks, have historically been one of the more successful means of accomplishing traffic-engineering goals. Switched-transport networks are based on a core of ATM switches connected to each other and to routers by logical links known as PVCs that live at Layer 2 of the OSI reference model. Figure 12-4 shows this type of topology.

Figure 12-4. Routers Connected by PVCs Ringing the Edge of an ATM Core

The routers that operate at the edge of the switched-transport network see each PVC as a point-to-point circuit to other distinct routers. These routers being OSI Layer 3 devices, they have a completely different view of the switched-transport network topology, as is shown in Figure 12-5.

Figure 12-5. The Routers' View of a Switched-Transport Network

After the routers are linked over the ATM core, the router's IGP is configured across each of the PVCs to establish peer relationships with the other routers and exchange routing information. A handy feature of the switched-transport solution to traffic engineering is its ability to handle link failure through the use of redundant PVCs. With redundant PVCs configured, IGP metrics can be used to ensure that backup PVCs are used when the primary PVC is not available. Also, if the primary PVC becomes available after an outage , traffic can automatically be sent back over the primary PVC, again through the metrics of the router's IGP.

Although it is a feasible and commonly used traffic-engineering mechanism, this method has the following pitfalls:

• The N -Squared Issue ” N -squared refers to a common simplification of the equation that is used to calculate how many connections are required to establish a full-mesh topology. The equation, ( N ( N “1))/2, where N is the number of routers in a network, shows that to establish a full-mesh topology in a network with 100 routers, 4,500 PVCs must be configured. Deploying a full mesh of PVCs stresses the IGP. This stress results from the number of IGP peer relationships that must be maintained by the router.

• ATM Cell Tax ”Every ATM cell is 53 bytes in length: 48 bytes are used for the cell payload and 5 bytes are used for the cell header. Therefore, ATM overhead is never less than about 10 percent, which means that when ATM is used, the ATM cell header consumes 10 percent of the available bandwidth. This wastes expensive WAN bandwidth.

• ATM SAR Speed ”Since ATM cells are used between the routers, and IP packets are used within the routers, each ATM router interface must have a SAR chip, which breaks IP packets into ATM cells as they leave the router on an ATM interface and reassembles ATM cells into IP packets when they arrive on a router's ATM interface. The problem is that ATM SAR processing speed has fallen behind the faster processing speeds provided by the SONET SAR chip. This means that ATM technology has not advanced enough to enable its use in higher-speed networks.

• Separate Autonomous Control Planes ”With switched-transport networks, there are two separate and autonomous control planes. In the first plane are the routers. The routers rely on IGP calculations to forward traffic. In the second plane are the switches. The switches rely on a provisioning tool to build PVCs. The IGP has no control over the switches. The PVC provisioning tool has no control over the routers.

• Additional Points of Failure ”Switched-transport networks require the use of complex ATM switches that require a high degree of competence to configure. The addition of these devices adds additional points of both human and mechanical failure to the network environment.

Because of these drawbacks, switched-transport networks are seldom the best solution for traffic engineering.

12.2.3 MPLS

The original purpose of MPLS was to make IP routers faster. It was perceived that if the router could be forced to function more like a switch with regards to making forwarding decisions based on OSI Layer 2 information, then it could move traffic as fast as a switch.

This was the perception in the mid 1990s of the primary purpose for MPLS. Since that time, routers have evolved to the point where they are capable of making forwarding decisions based on OSI Layer 3 information just as quickly as, if not more quickly than, a switch can make its Layer 2 “based forwarding decision. Some engineers still believe that the use of MPLS brings improvements over IP routing with regards to packet forwarding speed. This is simply not true.

MPLS has not been abandoned as an outdated mechanism because it is the best traffic-engineering tool available for IP networks. It can control traffic over numerous physical media types or across a network without interfering with native IGPs while remaining immune to their metrics-based decisions. MPLS offers the same level of control as seen with an ATM core without being burdened with traditional ATM drawbacks. It offers a solution to the complexity and expense of supporting two autonomous sets of equipment. MPLS is not automatically subjected to the bandwidth limitations of ATM SAR interfaces. And lastly, MPLS takes some of the burden off of the IGP, rather than stressing it the way ATM PVCs do.

The question as to why MPLS is good for traffic engineering has now been answered . The following sections discuss how MPLS works.