January 2002

Powering Core Networks Through MPLS

BY TONY RYBCZYNSKI AND BILEL JAMOUSSI

Most organizations have made Web-enabled e-Business a critical component of their go-to-market and operational models. The networked online economy � for all its volatility � is a fact of life. Service providers have been hard-pressed to keep pace. Too often, terms like �World Wide Wait,� and �road kill on the Information Highway� define the typical user experience... forced to wait and hope for access, then wait and hope for best-effort delivery. Multi-Protocol Label Switching (MPLS) is an important enabling technology for service providers. MPLS addresses the key problems faced by today�s IP service provider core networks � speed, scalability, quality of service (QoS), management, and traffic engineering � by providing a mechanism for efficient routing and resource reservation. While some larger enterprises may build their own MPLS networks, most would only see the economic and performance benefits in the Layer 1, 2, or 3 services they buy from service providers.

MPLS initially enhanced the performance and scalability of IP networks by replacing traditional IP packet forwarding � with its complicated address-matching algorithms � with fast, simple switching based on labels. This functionality enabled basic traffic engineering capabilities that, up to recently, were delivered by ATM. However, subsequent advances in ASIC design have given rise to a new generation of routing engines that are capable of forwarding IP packets at wire speed, reducing the requirement for MPLS strictly for performance improvement. So why is MPLS still of interest? The answer, in a word: Convergence. Data networking concepts are being applied in new domains such as optical, wireless, and voice networks and MPLS is at the heart of this convergence revolution.

WHY NETWORKS NEED MORE THAN IP
Wait-and-hope is the result of hop-by-hop � the fundamental paradigm of IP packet forwarding. Imagine for a moment that you�re a data packet trying to get from A to B. IP routing goes something like this: As you hop from router to router, continually ask which next hop will bring you closer to B. Go there, repeat, and keep doing that until you reach B. �Going to B? You�d better go to D, it�s on the way.� Never mind that D might be congested, or the link from D to B might be down, or the shortest path might not be the fastest. You�re still told that D is the best way to go, until you find out otherwise.

Since not all IP packets are created equal, some packets, such as those carrying voice and video for example, may not be able to reach their destinations promptly enough to meet application needs. They can get stuck behind other packets whose intrinsic QoS requirements are not as sensitive to transit delays brought about by this best-effort delivery model. This makes traditional hop-by-hop IP packet forwarding, by itself, ill-suited for large-scale use with revenue generating applications such as voice and video. This is because the choice of a routing path is not coupled with any knowledge of the necessary path characteristics required at the application level nor with any ability to find better suited routes based on knowing the overall topology and available resources of the network.

That�s where MPLS comes in. MPLS provides a way to optimize packet flows within IP networks through standards-based signaling, routing, and path computation intelligence. These capabilities enable the QoS mechanisms required for delivering profitable services, ensure efficient network resource utilization, and unify the network infrastructure across multiple protocols. Initially MPLS was multi-protocol in name only, intended to add a lot more order into IP networking through the definition of Layer 2 label switched connections. The importance of MPLS has grown as it has become multi-protocol above and below. Above, very high capacity MPLS switches will support not only IP packets, but also Ethernet frames, ATM cells, and bit streams. Below it will support not only optical pipes and ATM, but also switched wavelengths (through an extension called Generalized MPLS) and wireless. MPLS is therefore a key strategic convergent technology for core networks.

MPLS: THE 5,000-FOOT VIEW
MPLS works by adding labels to multiple types of payloads, switching these labeled packets at intermediate nodes and carrying these packets on whatever the underlying networking infrastructure is. IP, Ethernet, ATM, and bit streams can all be carried on a common MPLS core. Regardless of the underlying protocol, these payloads are forwarded using the MPLS label attached to them � a label that changes at every node along the way. There is no need to read all that detail in the originally assigned Layer 2 or Layer 3 headers. MPLS only needs to index a simple label to forward the packet on its way.

Imagine once again that you�re that packet trying to get from A to B. MPLS routing is like having friends go to B ahead of you, using hop-by-hop or source routing. For every road, they reserve a lane just for you. At every intersection, they post a big sign that indicates which way you should turn and what new lane to take to optimize your trip. Now you can get where you want to go faster, because there�s less traffic going your way and no unexpected delays. You don�t have to constantly tell people where you�re going (your full IP address) because they know from one glance at your MPLS label who you are and where you are headed. As a result, you get the service you need and your trip becomes much more predictable.

The route that is reserved for you is called a �label switched path� or LSP. In many respects, LSPs are just like switched paths in ATM or frame relay networks, except that MPLS paths are not bound to any particular Layer 2 technology. MPLS achieves this versatility by separating the control function using MPLS signaling and routing protocols originally designed for IP � which scout out the optimal route � from the forwarding function, which switches the data along the pre-assigned path. Since the control function doesn�t care what type of traffic it is controlling, an MPLS node can forward packets, frames, and cells with equal ease. Consequently, many existing ATM switches and IP routers can be re-purposed into MPLS nodes with a software upgrade to their control planes. No need to change out hardware-based forwarding functions.

This separation of control and forwarding enables MPLS to provide consistency across multiple core technologies. Imagine again that packet trying to get from Point A to B. It can originate on an Optical Ethernet MAN, then be switched across a gigabit MPLS core network, then across an optical cross-connect network, then onto an ATM backbone to its destination. MPLS can manage its travel across all these technologies, producing predictable performance and preserving investment in existing networking equipment. All of this would be transparent to the enterprise user, who is subscribing to a particular service (e.g., IP VPN, virtual private ethernet or frame relay or ATM).

TRAFFIC ENGINEERING AND QoS THROUGH MPLS NETWORKS
Basic traffic engineering is implicit in MPLS, because label switched paths are created based on service needs and current network information. The ingress node, the first MPLS-enabled node encountered by the user packet, can calculate the optimum path based on traffic requirements and network conditions. Updated info about network conditions are regularly broadcast, automatically or on demand, among MPLS nodes using OSPF (Open Shortest Path First) or IS-IS (Intermediate System to Intermediate System), two routing protocols commonly used in IP networks. Both of these control protocols have been enhanced with traffic engineering extensions for MPLS, these known as OSPF-TE and IS-IS-TE.
Information about MPLS labels � what they mean and what to do with them � is shared among network nodes using an MPLS signaling protocol called the Label Distribution Protocol (LDP). The basic Label Distribution Protocol is very effective for fast, packet forwarding, for streamlining routing tables on intermediate nodes, and for creating VPN tunnels, but it�s not particularly useful for traffic engineering.

For dynamic QoS-based traffic engineering, the ingress MPLS node uses enhanced MPLS control protocols to identify and reserve the best label-switched path, to support a range of traffic engineering mechanisms, including:

• Strict/explicit routing, whereby every intermediate node on the path is specified in advance;
• Loose routing, whereby some segments of the path are specified in advance and others are left unspecified for now, open to re-routing around network problems, if necessary;
• Pre-emption, whereby more-critical paths take precedence over less-critical paths, as defined by eight priority levels; and
• Route pinning, which specifies whether or not a route can be modified or pre-empted.

As sometimes happens through the standards process, two MPLS control protocols have been defined:

• CR-LDP (Constraint-based routing Label Distribution Protocol) is an extension to the basic LDP. Unlike basic LDP, CR-LDP supports explicit routing parameters. For example: �Let the path accept up to this amount of bandwidth, up to this peak rate�; �Only chart this path onto nodes that guarantee this level of service�; �When allocating resources, consider this path more important than those other paths�; �Don�t set up a route that takes more than 15 hops�; or �Consider this path-calculation parameter more negotiable than those other parameters.�
• RSVP-TE (Resource Reservation Protocol with traffic engineering) is an extension to the decade-old RSVP (Resource reSerVation Protocol), and supports capabilities that are somewhat similar to CR-LDP. RSVP-TE supports a narrower set of service classes: �I�ll do my best, and we�ll see what happens�; �I�ll do a good job, but no hard-and-fast guarantees�; and �I promise to give you all the service you need to stay within your expectations.�

MPLS supports the same QoS features as IP. Since MPLS supports reservation of Layer 2 resources, it also delivers finely grained QoS, much the same as ATM. MPLS also supports a range of queuing mechanisms for QoS differentiation between LSPs. One of the great values of MPLS is its ability to maintain these specific performance characteristics across any type of transport medium, eliminating the need for overlay networks and multiple control mechanisms.

JUST THE BEGINNING
Now, the control mechanisms that make MPLS so useful in the packet domain are being applied in the optical domain with a next-generation MPLS, known as Generalized MPLS (GMPLS). This extended version of MPLS enables devices in both the packet and optical layers to establish optical paths on demand, optimize resources, and share intelligence, which in turn sets the stage for profitable optical services such as optical VPNs, storage area networks, and bandwidth trading.

MPLS and GMPLS redefine packet and optical networks, enabling the introduction of new services together with enhanced methods to deliver existing services more efficiently, while simultaneously extending the value of existing network equipment. By integrating these protocols into core networks, service providers can provide an intelligent, high-performance Internet to its enterprise and consumer customers, while recognizing revenue opportunities with reduced operational costs. 

Tony Rybczynski is director of strategic marketing and technologies for Nortel Networks� Enterprise Solutions unit. E-mail questions or comments to [email protected]. Bilel Jamoussi is chief standards strategist at Nortel Networks. E-mail questions or comments to [email protected]. More information can be found on the company�s Web site: www.nortelnetworks.com.

[ Return To The January 2002 Table Of Contents ]