- What Is MPLS?
- Why Is MPLS Needed?
- How Is MPLS Done?
- When and Where Is MPLS Used?
- Who Is Doing MPLS?
- The Label Switching Paradigm
- A Quick Introduction to MPLS
- Evolution of Internet Network Models
- Basics of the Internet
- Internetworking Technology Basics
- More Basics: Graph Theory and Modeling Language
- The Promise of MPLS
- The Promise of the Promise of MPLS
Evolution of Internet Network Models
The need for MPLS can be traced through the recent evolution of the Internet by looking at several network models that were developed to handle the routing of data.
There are four models that are of interest:
- The IP over ATM overlay model
- The multilayer switching model
- The "pure IP" model
- The MPLS-only model
In the early-to-mid 1990s, ATM became a popular solution for providing transport services within service provider networks. The ATM speeds of up to 622 Mbps were greater than the TDM T-1 and T-3 speeds that were currently in use in the Internet. Since higher level applications were always primarily IP-basedthere were very few "native" ATM applications ever developedthe IP over ATM overlay model was created as a reliable and cost-effective way to multiplex IP data over an ATM core. The model is presented in Figure 112.
Figure 112 The IP over ATM overlay model.
The main feature of an overlay network is that the model presents two independent networks: a Layer 2 network that is running ATM in the core, and a Layer 3 network that is IP-based and running at the edges of the networks. Thus, at the IP layer, the IP devices are only logically connected from edge to edge. The IP and ATM protocols work concurrently, but in a totally independent fashion; the phrase "ships in the night" is often used to refer to this configuration.
There are several advantages of the overlay model. When VCs are set up in ATM, TE and QoS parameters can be specified because of the connection-oriented nature of the protocol. Protection and failover are often provided by an underlying SONET physical network in these types of networks, also. The ATM switches are arranged in a full-mesh configuration to guarantee the any-to-any connectivity that is required from the routed edge traffic's point of view. At the time, this arrangement provided state-of-the-art transport for service providers. It is still in use today with a large installed base.
There are also several disadvantages of the model. The full-mesh topology of the ATM switch core does not scale well when new switches or edge routers need to be added (this is called the "n-squared" problem). The full-mesh configuration can also cause a routing convergence problem for Internal Gateway Protocols (IGPs) during a link failure because of the number of updates that need to be sent. Because the Layer 2 and Layer 3 networks are independent, they must be managed and administered by two completely separate management platforms. This is a complex and costly situation. Also, ATM has a problem scaling past the OC-48 speed because it is not cost-effective to create the required segmentation and reassembly (SAR) chips to do the cell assembly and reassembly functions. Finally, because ATM is based on 53-byte fixed cells, there is a "cell tax" when transporting variable-length-packet IP traffic. Some of the data payloads are empty. There is work going on in developing a longer, frame-based ATM, but it has yet to be seen if this effort will have a place in the current internetworking landscape.
Multilayer Switching Networks
In the mid-1990s, many innovative ideas were proposed to bring the best features of the connection-based, VC, label swapping ATM and FR technologies to the Internet. Vendors developed innovativealbeit not standard, and therefore non-interoperablesolutions for the market. Such concepts as IP switching, the cell-based router, and products from Cisco, IBM, Ascend (Lucent), and others were attempts to integrate the overlay model of ATM and IP by combining the Layer 3 IP control plane with the Layer 2 ATM label swapping forwarding capabilities. This multilayer switching model is shown in Figure 113.
Figure 113 Multilayer switching model.
With multilayer IP switching, the IP routing control process is used with ATM label handling for data forwarding of the ATM cells carrying the packet traffic in the data payload. The ATM control plane protocols are not used. In this model, there are no longer "ships in the night," but rather "one boat." There is only one management and administration of the network, making it simpler and cheaper than the overlay model. Another big advantage is that it uses standard IP addressing. IP services and applications run natively on the network.
The big gotcha, however, is that these solutions were never standardized; therefore, the devices could not work with each other. Several important desirable ATM and FR features were also lost, including QoS and the ability to do TE and management.
"Pure IP" Networks
IP has become the predominant network layer protocol in use today, and the evolution to a "pure IP" network seems a natural direction, especially with the development of IP-based routing at very high speeds. This type of model is shown in Figure 114. Both the control and data forwarding planes are based solely on IP technologies.
Figure 114 "Pure IP" model.
This is the current direction in the Internet core as IP-based terabit routers now begin to outperform ATM switches. New optical technologies are allowing for the creation of OC-192 and OC-768 speeds in these devices. These devices add an even greater dimension of simplicity and ease of management. The introduction of new standardized technologies such as MPLS will also add the missing features from the ATM protocol. Required applications such as QoS and TE, plus the ability to have deterministic performance when required by using connection-like paths and the label swapping data forwarding plane, will be added to the network.
If the ideas of a universal control plane and label swapping data forwarding plane are advanced even further, a case may be made for MPLS-only networks as the next natural evolutionary step for certain classes of network devices within the Internet. Figure 115 shows two communicating LSRs that exclusively use the MPLS control and forwarding planes:
Figure 115 The MPLS-only model.
This scenario is currently not feasible because MPLS is tightly coupled with other TCP/IP protocols. The MPLS domain sits "inside" a larger IP domain to provide the necessary end-to-end connectivity that is required for host-to-host communications with the Internet. If deployment of MPLS becomes ubiquitous, MPLS-only networks may evolve for specific purposes, such as "fat" tunnels for moving large amounts of special data, many new evolving optical applications, and other uses that have not been developedor imaginedyet. An MPLS-only device is a current development topic, and such development may produce devices that fill a specific niche in some networking situations.
To come full circle after looking at the future of the Internet, it is beneficial to look at the basics of the Internet to see how we got here in the first place.