Multi-Protocol Label Switching (MPLS)

Today the division between routers and switches is a fine line. Whereas switches were initially designed to help segment a LAN into multiple collision domains, and thereby allow you to extend the reach of a particular LAN topology, switches have moved higher up the networking ladder. When switching is used in a LAN to connect individual client and server computers, the process is known as microsegmentation, because the collision domain has been reduced to just the switch and the computer attached to a port. Switches at this level generally work using the hardware (MAC) addresses of the attached computers.

Layer 3 switching moves switching up the ladder by one rung by switching network frames based on the OSI Network layer addressan IP address, for example. But wait, that's what a router does, isn't it? Of course. A layer 3 switch is basically a router, but it implements most of its functions in application-specific integrated chips (ASICS) and performs its packet processing much faster than does a traditional router, which uses a microprocessor (much like a computer CPU) for this function.

Details on layer 3 switching can be found in Chapter 8, "Network Switches."

When you get to the top of the ladder, where large volumes of data need to be routed through a large corporate networkor the Internet, for that mattereven the fastest traditional routers or layer 3 switches easily can become bogged down by the volume of traffic. Because of this, the core of a large network traditionally has been built using ATM or Frame Relay switches, and IP traffic is sent over these switched networks.

To speed up the processing of routing packets at high-volume rates, a newer technology has been developing over the past few years and goes by the name of Multi-Protocol Label Switching (MPLS). MPLS is covered in the next section.

Combining Routing and Switching

Traditional routers have a large amount of overhead processing they must perform to get a packet to its destination. Each router along the packet's path must open up and examine the layer 3 header information before it can decide on which port to output the packet to send it to its next hop on its journey. If a packet passes through more than just a few routers, that's a lot of processing time. Remember that IP is a connectionless protocol. Decisions must be made about a packet's travel plans at each stage of its journey through the network. The solution to this problem lies in newer technologyhigh-speed switching. Specifically, Multi-Protocol Label Switching, which is discussed in the next section, combines the best of routing techniques with switching techniques.

When you look at concepts such as ATM or Frame Relay, which are connection-oriented protocols, this isn't the case. Instead, virtual circuits (either permanent or switched) are set up to connect to endpoints of a communication path so that all cells (as in the case of ATM) or frames (as in the case of Frame Relay) usually take the same path through the switched network.

For more information about ATM and Frame Relay and how these connection-oriented switched networks function, see Chapter 15, "Dedicated Connections."

Adding a Label

MPLS is a method that takes the best of both worlds and creates a concept that allows IP packets to travel through the network as if IP were a connection-oriented protocol (which it isn't). Using special routers called Label Switching Routers (LSRs) does this. These routers connect a traditional IP network to an MPLS network. A packet enters the MPLS network through an ingress LSR, which attaches a label to the packet, and exits the MPLS switched network through an egress LSR. The ingress LSR is the router that performs the necessary processing to determine the path a packet will need to take through the switched network. This can be done using traditional routing protocols such as OSPF. The path is identified by the label that the ingress router attaches to the packet. As you can see, the ingress router must perform the traditional role that a router fills. It must perform a lookup in the routing table and decide to which network the packet needs to be sent for eventual delivery to the host computer.

However, as the packet passes through the switched network, it is only necessary for the switch to take a quick look at the label to make a decision on which port to output the packet. A table called the Label Information Base (LIB) is used in a manner similar to a routing table to determine the correct port based on the packet's label information. The switch doesn't perform IP header processing, looking at the IP address, the TTL value, and so on. It just spends a small amount of time doing a lookup of the label in the table and outputting the packet on the correct port.

Note

Similar to ATM and Frame Relay networks, the label attached in an MPLS network doesn't stay the same as the packet travels through the network of switches. Labels are significant locally and only identify links between individual switches. Depending on the MPLS implementation, labels can be set up manually (like permanent virtual circuits) or can be created on-the-fly (like switched virtual circuits). However, after a path (or circuit) through the switched network has been created, label processing takes only a small amount of time and is much faster than traditional IP routing. Each switch simply looks up the label, finds the correct port to output the packet, replaces the label with one that is significant to the output port, and sends it on its way.

When the packet reaches the egress LSR, the label is removed by the router, and then the IP packet is processed in the normal manner by traditional routers on the destination network.

If this sounds like a simple concept, that's because it is. MPLS still is in the development stages, so you'll find that different vendors implement it in different ways. Several Internet draft documents attempt to create a standard for MPLS. Other features, such as Quality of Service (QoS) and traffic management techniques, are being developed to make MPLS a long-term solution.

Using Frame Relay and ATM with MPLS

One of the best features about the current design of MPLS is that it separates the label-switching concept from the underlying technology. That is, you don't have to build special switches that are meant for just MPLS networks. MPLS doesn't care what the underlying transport is. It is concerned only with setting up a path and reducing the amount of processing a packet takes as it travels through the circuit.

Because of this, it's a simple matter for an ATM or Frame Relay switch vendor to reprogram or upgrade its product line to use MPLS. For ATM switches, the VPI (virtual path identifier) and VCI (virtual channel identifier) fields in the ATM cell are used for the Label field. In Frame Relay switches, an extra field is added to the IP header to store the label. However, don't get confused and think that an MPLS network is an ATM network or a Frame Relay network. These switches must be reprogrammed to understand the label concept. It's even possible, for example, for an ATM switch to switch both ATM and MPLS traffic at the same time. By allowing for the continued use of existing equipment (and these switches are not inexpensive items), large ISPs or network providers can leverage their current investment, while preparing to install newer MPLS equipment when the standards evolve to a stage that makes it a good investment.

For the long term it's most likely that MPLS will be implemented using technology similar to Frame Relay instead of ATM. This is because of the small cell size of the ATM cell (53 bytes) combined with a high overhead (the 5-byte cell header). In a small network with little traffic, this 5-byte cell header seems insignificant. However, when you scale this to large bandwidth network pipes, this amount of overhead consumes a large amount of bandwidth given the small amount of data carried in the 53-byte cell. Thus, variable-length frames are most likely to become the basis for MPLS networks in the next few years.