Mul­ti­Pro­to­col Label Switching (MPLS)

There are basically two types of data transfers: With con­nec­tion­less data transfer, data is sent at any time and without limits from the desired device to a target system without the path of the packages being set in advance. Each in­ter­con­nect­ed network node (usually a router) au­to­mat­i­cal­ly knows how to forward the data stream. Con­nec­tion­less data transfer offers a high degree of flex­i­bil­i­ty, but no guarantee that the necessary resources will be available.

In contrast, the path of the data packages in con­nec­tion-oriented data transfer is decided from the beginning. The network nodes involved (generally switches) receive the cor­re­spond­ing in­for­ma­tion for the for­ward­ing of the data from the preceding station, until the point when the packages have arrived at the target computer at the end of their path. In this way, the time-consuming routing process necessary for a con­nec­tion­less transfer is con­sid­er­ably ac­cel­er­at­ed. It also allows for optimal control of available network resources and their dis­tri­b­u­tion amongst in­di­vid­ual par­tic­i­pants. The so-called mul­ti­pro­to­col label switching (MPLS) allows this to be used for TCP/IP networks as well, though these are tech­ni­cal­ly in the con­nec­tion­less network category.

In the mid-1990s, large com­mu­ni­ca­tion networks were char­ac­ter­ized by a much higher pro­por­tion of voice com­mu­ni­ca­tion (telephony) than data com­mu­ni­ca­tion (internet). Telecom­mu­ni­ca­tion providers at this time were still operating separate networks for both transfer types, which was quite expensive, but also didn’t ensure a com­pre­hen­sive quality of service. The high quality, con­nec­tion-oriented com­mu­ni­ca­tion networks stood in contrast to con­nec­tion­less data networks, which lacked the necessary bandwidth. The in­tro­duc­tion of the ATM protocol (Asyn­chro­nous Transfer Mode) largely solved this problem by allowing the voice and data to be trans­ferred via a shared in­fra­struc­ture. But mul­ti­pro­to­col label switching really provided a solution in the late 1990s for using available band­widths ef­fi­cient­ly.

To ac­com­plish this, MPLS finally relieved the over­loaded routing systems: Instead of defining the optimal route of a data package by the in­di­vid­ual in­ter­me­di­ate stations, as was done before, the new method offered the pos­si­bil­i­ty to pre-define paths that define a package’s way from the input point (ingress router) to the starting point (egress router). The relay points (label switched router) recognize these paths by eval­u­at­ing labels that contain the ap­pro­pri­ate routing and service in­for­ma­tion, and are assigned to the re­spec­tive data package. The eval­u­a­tion takes place using the ap­pro­pri­ate hardware (e.g. a switch) above the backup layer (layer 2), while the time-consuming routing on the switching layer (layer 3) is left out.

Thanks to the gen­er­al­ized MPLS extension, the tech­nol­o­gy orig­i­nal­ly developed only for IP networks is now also available for other network types, such as SONET/SDH (Syn­chro­nous Optical Net­work­ing / Syn­chro­nous Digital Hierarchy) or WSON (Wave­length Switched Optical Network).

The use of MPLS in IP networks requires a logical and physical in­fra­struc­ture con­sist­ing of MPLS-capable routers. The labeling process operates primarily within an au­tonomous system (AS) – a con­gre­ga­tion of different IP networks that are managed as a unit and connected via at least one common interior gateway protocol (IGP). Ad­min­is­tra­tors of such systems are generally internet providers, uni­ver­si­ties, or in­ter­na­tion­al companies.

Before the in­di­vid­ual paths can be built, the IGP being used needs to ensure that all routers of the au­tonomous system can reach one another. Then the corner points of the paths, which are also referred to as label switched paths (LSP), are defined. These pre­vi­ous­ly mentioned ingress and egress routers are usually at the inputs and outputs of a system. Ac­ti­va­tion of the LSPs is then either manual, semi-automatic, or fully automatic:

• Manual con­fig­u­ra­tion: Each note that an LSP runs through needs to be in­di­vid­u­al­ly con­fig­ured; this approach is in­ef­fec­tive for large networks.

• Semi-automatic con­fig­u­ra­tion: Only some in­ter­me­di­ate stations (for example, the first three hops) need to be con­fig­ured manually, while the rest of the LSPs receive in­for­ma­tion from the interior gateway protocol.

• Fully automatic con­fig­u­ra­tion: The interior gateway protocol assumes the entire de­ter­mi­na­tion of the path in the fully automatic version; no path op­ti­miza­tion is achieved, though.

Data packages sent in a con­fig­ured MPLS network receive an ad­di­tion­al MPLS header from the ingress router. This is inserted between the in­for­ma­tion of the second and third layers, and is also referred to as a push operation. During the transfer, the in­di­vid­ual hops involved exchange the label with a cus­tomized version with its own con­nec­tion in­for­ma­tion (i.e. latency, bandwidth, and des­ti­na­tion hop) – this procedure is often called a swap operation. At the end of the path, the label is removed from the IP header as part of a pop operation.

The ad­di­tion­al 32 bits of the MPLS label stack entry add four pieces of in­for­ma­tion to an IP package for the next network hop:

• Label: The label contains the core in­for­ma­tion of the MPLS entry, which makes it the largest component with a length of 20 bits. As mentioned before, a label on the path is always unique, and only mediates between two specific routers. It’s then adapted ac­cord­ing­ly for the data transfer to the next in­ter­me­di­ate station.

• Traffic Class (TC): Using the traffic class field, the header provides in­for­ma­tion about dif­fer­en­ti­at­ed services (DiffServ). This formula can be used for the clas­si­fi­ca­tion of IP packages to guarantee service quality. For example, the 3 bits of the network scheduler can com­mu­ni­cate whether a data package is pri­or­i­tized or can be sub­or­di­nat­ed.

• Bottom of Stack (S): The bottom of stack defines whether the un­der­ly­ing trans­mis­sion path is a simple path or whether multiple LSPs are nested. If the latter is the case, then a package can receive multiple labels grouped together in the so-called label stack. The bottom of stack flag then informs the router that other labels are following, or that the entry contains the last MPLS label in the stack.

• Time to Live (TTL): The last 8 bits of the MPLS label stack entry shows the lifespan of the data package. In this way, it’s possible to control how many routers the package can go through on its path (the limit is 255 routers).

In the 1990s, MPLS helped providers with the rapid de­vel­op­ment and growth of their networks. But the initial speed advantage for data transfer has been pushed to the back­ground with the new gen­er­a­tion of high-per­for­mance routers with in­te­grat­ed network proces­sors. As a procedure that can guarantee quality of service, though, it’s still used today by many service providers. This has to do with so-called traffic en­gi­neer­ing – a process that deals with the analysis and op­ti­miza­tion of data streams. In addition to the clas­si­fi­ca­tion of in­di­vid­ual data con­nec­tions, an analysis of the band­widths and ca­pac­i­ties of in­di­vid­ual network elements also takes place. Based on the results, the data load can then be optimally dis­trib­uted to strength­en the entire network. Another important area of ap­pli­ca­tion is virtual private networks (VPN) – self-contained, virtual com­mu­ni­ca­tion networks that use public network in­fra­struc­tures like the internet as a transport medium. In this way, devices can merge in a network without phys­i­cal­ly con­nect­ing with one another. There are two different types of such MPLS networks:

• Layer 2 VPNs: Virtual private networks on the data link layer can either be designed for point-to-point con­nec­tions or for remote access. Layer 2 logically serves the user of such a VPN as an interface for es­tab­lish­ing a con­nec­tion. The point-to-point tunneling protocol (PPTP) or the layer 2 tunneling protocol (L2TP) serve as basic protocols. This gives service providers the option to offer their customers SDH-like services and Ethernet.

• Layer 3 VPNs: Network-based layer 3 VPNs represent a simple solution for service providers to offer various customers com­plete­ly routed network struc­tures based on a single IP in­fra­struc­ture (re­gard­less of the private IP address ranges). The quality of service is ensured by way of customers being managed sep­a­rate­ly by in­di­vid­ual MPLS labels and pre­de­fined package paths. The network hops are also spared routing.

Operators of large WAN networks (Wide Area Network) profit from provider offers that are based on mul­ti­pro­to­col label switching: Correctly con­fig­ured, the strategic label switched paths optimize data traffic and ensure to the greatest extent possible that all users can have the bandwidth that they need at any time – while their own effort remains limited. For campus networks, such as uni­ver­si­ty or en­ter­prise networks, the method is also a suitable solution, if the necessary budget is available.