November 2000

Rearchitecting IP Networks

BY TONY RYBCZYNSKI

The state-of-the-art is a hierarchical network of routers connected by high-speed links running directly on optical systems or on ATM networks. These networks are complex to engineer and operate, take a long time to recover from failures, and support only best-effort IP traffic.

All this is about to change, however, with MPLS, which provides a transition path from today's public multiple-network world to a single label-switched network running on smart optical wavelengths. While driven by the needs of IP networking, MPLS will also ultimately support circuit emulation (dedicated bandwidth pipes) and switching of frame relay and ATM virtual circuits. In enterprise networks, it provides a glue and transition path from today's multiservice-on-ATM backbones to highly reliable application-aware IP networks running directly on optics. A significant difference between MPLS and the vision of ATM in the early 90's is that MPLS is optimized for the ultimate world of all applications running on IP, while allowing network operators to support dedicated bandwidth (Layer 1), virtual bandwidth (Layer 2), and IP-based connectivity.

LABELS FOR TRAFFIC MANAGEMENT
What's driving MPLS? Fundamentally, MPLS is driven by the need to improve the traffic management of large-scale IP networks supporting multiple traffic classes and intrinsic virtual private network (VPN) operation. Traffic management is all about making optimum use of network resources under unpredictable traffic loads and failure conditions while meeting application needs.

An analogy here would be instructive. Consider traffic management for an expressway, such as that coursing through a metropolitan center. This kind of traffic management is geared towards keeping the traffic flowing as close to the theoretical capacity as possible, while meeting the needs of various types of vehicular users (buses, commuters, emergency vehicles). If traffic management capabilities are inadequate, traffic overloads result which may bring the expressway to a standstill. Similarly, traffic overloads and inadequate traffic management capabilities are a significant contributor to IP network failures and performance deficiencies.

Flexible multiservice edge switches and high capacity MPLS core switches (referred to in the standard as Label Edge and Label Switch Routers, respectively) riding on top of smart optics will finally allow IP networks to exhibit the reliability, performance, and operational simplicity required to support the broadest range of applications and services. Unfortunately, MPLS doesn't apply to asphalt-based networks!

PRECURSORS TO LABELS
The Internet and enterprise IP networks have made a success of conventional connectionless IP networking. Minimizing the amount of state information maintained in the network has resulted in scalability, flexibility, and cost-effectiveness that has been demonstrated over the years. The ability of IP to ride over virtually any transmission technology has also contributed to its success.

In recent years, network engineers have found that transporting connectionless IP on top of connection-based Layer 2 networks provided them with capabilities not found in IP router networks. One such capability -- making good use of bandwidth capacity under changing traffic patterns -- has always been problematic with IP networks. To overcome this limitation, network engineers initially relied on frame relay and, more recently, on ATM, and in this way took advantage of the availability of high-speed, low-cost switches and matching interfaces for routers.

The advantage of riding IP on top of ATM is the ability to do traffic engineering using ATM's connection-oriented capabilities and to leverage ATM's class-of-service capabilities. This also goes some way towards improving the reliability of IP networks, lowering the packet loss rate and reducing delay variation. Finally, ATM allowed IP traffic to be handled on an integrated network, also supporting conventional voice, data, and video traffic.

More recently, engineers started to look at how connection-oriented capabilities could be introduced into IP networks that could not only simplify operation over ATM networks but also provide an evolution path to packet-on-SONET and packet-on-wavelength operation. As the work progressed, it was recognized that the same techniques that could enhance traffic management of IP networks could also enable support for large-scale IP class-of-service and VPN capabilities. Running IP over ATM was a precursor that demonstrated the value of leveraging connections to improve the performance and manageability of IP networks. MPLS goes further by eliminating the need for two separate signaling planes (IP and ATM) while allowing both to coexist in a single network.

SWITCHING LABELS NOT PACKETS
At the highest level, MPLS defines an architecture whereby labels are assigned to packets at the edge of the network (these labels being used at tandem switches to route the packet to the destination). This approach avoids the need to analyze the IP address fields of every packet. MPLS integrates the label-swapping paradigm of Layer 2 technologies (such as ATM or Frame Relay) and the signaling functionality of connection-oriented networking with the routing paradigm of Layer 3 protocols such as IP.

Since labels can be assigned at the edge of the network on a broad range of attributes -- such as source/destination addresses, class of service, and user and security classes -- label switching represents a different routing paradigm for IP networking, very different from packet-by-packet, hop-by-hop operation. More specifically, MPLS is designed to:

• Eliminate the need to look up IP addresses in every node along the path.
• Provide a means of explicitly engineered routes and flows to balance traffic across the network.
• Provide a means of explicitly engineering routes to create specialized services, differentiated by their ability to act as tunnels, offer quality of service (QoS), or both.

The MPLS architecture operates at three interrelated levels or planes: the data plane, the route determination plane, and the signaling plane. The data plane governs movement of user data across the network along label-switched paths. Each label switch router in the network identifies the incoming label and then looks up a table to identify the egress label and forwarding path for the packet. MPLS enhances the traffic management of connectionless networks by effectively codifying the current topology and forwarding paradigms into the form of a label-swapping table. To set up label-switched paths, the network needs to know where the destination is (route determination plane) and create a route that is known by all along the path (signaling plane).

The route determination plane maintains a routing view based on network topology, traffic patterns, and application needs that are used to define label-route associations. Layer 3 routing protocols such as Open Shortest Path First (OSPF) or Boundary Gateway Protocol (BGP) are used for discovery of the network topology. For handling label path determination, there are three options:

• It can be topology-driven and performed on a dynamic hop-by-hop basis following the route determined in the routing plane.
• It can be configuration-driven, still following the route determined in the routing plane.
• It can be constraint-based, in which case provisioning is used to set up a label-switched path along a provisioned route. In this case, label-switched paths are set up based on a set of attributes or constraints such as QoS or membership in a VPN.

Constraint-based routing represents the crown jewels of MPLS, providing a way to establish paths that satisfy certain QoS parameters, setup and holding preemption priorities, and VPNs. In addition, these paths may disperse traffic to optimize link utilization. Thus, constraint-based routing replaces repeated metric manipulation in connectionless networks and obviates ATM virtual circuit configuration complexity.

Enhancements to existing routing protocols (that is, OSPF) will be used to determine the best paths upon setup, in the event of link failures, or during periods of increased traffic. Network operators could maintain a model of their network offline to optimize the network links and better predict network behavior. This model imports traffic statistics from the network, analyzes the information, and then periodically updates network nodes to optimize link utilization. Given that forecasting traffic demands and communities of interest is virtually impossible in IP network environments, comprehensive traffic management achieved through topology-driven, constraint-based routing is a key value of MPLS.

In the signaling plane, a label distribution protocol (LDP) has been defined to ensure consistent label handling at transit nodes. There are currently two protocols being developed to support the signaling requirements of constraint-based routing: one approach defines enhancements to LDP with added flow specification capabilities; the second approach is based on redefining the ReSource reserVation Protocol (RSVP) for this mode of operation.

The experts note that both approaches have various advantages and disadvantages. Relevant considerations include signaling overhead, state management, behavior under failure conditions, resilience, scaling of total explicit routes, and tunnel types (such as point-to-point and point-to-multipoint). Several vendors (including Nortel Networks, which championed the former approach) are committed to supporting both methods in their products and interworking between the two. MPLS also defines a number of implementation options that pertain to label assignment, failure recovery modes for label-switched paths, merging of label-switched paths, and handling of class-of-services within label-switched paths.

THE VALUE OF A LABEL
The advantages of MPLS for large enterprises, such as financial institutions, manufacturers, and utilities, as well as for service providers, include higher reliability through better traffic engineering, more consistent application performance, and multiprotocol above and below for future-proof networking. MPLS can run on virtually any Layer 1 or 2 networking infrastructure, including ATM, and is extendible, in the sense that it may carry other protocols and traffic types in the future.

MPLS provides a sound networking infrastructure to meet application performance needs, including connectivity, throughput, latency, and security, through its ability to flexibly assign labels based on a broad range of user/application attributes. MPLS is the future-proofing glue for IP networking, allowing network operators to evolve from ATM to optical networking.

If you want the Internet to be as reliable, consistent, and secure as the telephone network, then MPLS is a key enabling technology whether you are a large enterprise or a service provider.

Tony Rybczynski is director of strategic marketing and technologies for Nortel Networks' Enterprise Solutions unit. E-mail questions or comments to [email protected]. 

[ Return To The November 2000 Table Of Contents ]