November 2000

BY MARK VEIL

Extreme competitive business pressures and an insatiable consumer appetite for anytime, anywhere access to content and services are driving phenomenal advancements and innovation in networking and computing technology. This is particularly true in the Internet protocol (IP) application space. Today we work, live, and play on the public network -- e-business, streaming video, and voice-over-packet applications are driving the convergence of multimedia applications to a common, ubiquitous IP-based platform.

Delivering converged, IP-based communications requires a new breed of access network, one that is not only engineered to deliver carrier-class service, but also optimized for today's packet-based services. This network must economically deliver these new services with the same, or better, quality than the existing infrastructure. More important it must be "service aware" -- capable of applying differentiated treatment and quality of service to traffic based upon the specific requirements of the applications being delivered. The ultimate objective is to manage, monitor, and control network traffic at the level of the service. These requirements necessitate the implementation of robust traffic management and traffic engineering services.

THE IP QoS CHALLENGE
Traditional IP networks operate on a connectionless, "best-effort" basis, with all packets subject to equal treatment as they are individually routed through the network on a hop-by-hop basis to their destination. Packets from the same flow may traverse the network over different paths, arrive at their destination out of sequence, and have to be reordered. Additionally, some of these packets will be lost in transit and have to be re-transmitted, and contention for resources and network processing and encoding overhead will slow the packet's journey.

These factors not only produce a cumulative delay, but they also introduce an element of unpredictability that manifests itself as delay variation. Moreover, IP's best-effort "fairness" translates to a relative "unfairness" for traffic that is more sensitive to network impairments. In times of heavy and/or prolonged network congestion, such impairments would likely produce some irritation for the Napster enthusiast experiencing longer-than-usual download times. However, for the business engaged in mission-critical B2B transactions, or relying upon packet-based voice services, the impact and repercussions are much more severe.

DELIVERING IP-BASED QOS
Quality of service (QoS) can be defined as the collective measure of service levels delivered to the customer premise and is characterized by its intrinsic behavioral properties and performance requirements. Delivering QoS, and meeting customer-contracted Service Level Agreement (SLA) obligations, requires the ability to manage and control the relevant service performance attributes -- attributes such as latency, jitter, average and peak packet rate, and packet loss ratios. By doing so, we can ensure that availability and performance is delivered within acceptable or contracted service bounds, and that premium or priority services are given preferential service within the network.

In the following sections we will examine traffic management and traffic engineering concepts and emerging solutions for delivering QoS in IP-based converged networks.

TRAFFIC MANAGEMENT
Traffic management is concerned with satisfying QoS performance objectives, for both new and existing traffic flows, and protecting against conditions that result in congestion and degradation of network performance. For traffic management to achieve its objectives, network elements must provide facilities for packet marking, traffic classification, admission control, and traffic shaping/conditioning. We'll consider each of these functions:

• Packet Marking -- Packets are annotated for a specific QoS treatment, such as queuing priority or drop precedence.
• Packet Classification -- Classifiers map packets requiring the same QoS requirements to specific outbound queues. Typically, these traffic classifications are based upon the contents of the packet header, such as the L2 and L3 source/destination address. In practice, however, classifications may be derived from (and applied to) a virtually unlimited range, combination, and granularity of packet attributes -- including physical ingress port/interface, application protocol type, IPv4 Type of Service (ToS), and IPv6 class of service (CoS) markings.
• Admission Control -- This function ensures that the requested traffic profile and QoS levels can be met with respect to current network state, resource availability, or other policy-based considerations prior to admitting the traffic flow.
• Traffic Shaping And Conditioning -- A variety of mechanisms are used to monitor and maintain compliance with traffic profiles (or traffic contracts). Metering services will monitor and measure traffic against its profile, and pass packets along to the appropriate policing mechanisms -- the queuing and dropping services.

For delivering converged services within the resource-constrained access network, it is generally accepted that traffic management services implement a fine granularity of control. The IETF's Integrated Services Model (IntServ) is well suited for this role by supporting explicit service guarantees for priority services, providing admission control based upon Resource ReSerVation (another IETF initiative), and delivering flow-level control (at the level of the service).

Within the core of the network -- where the issue is not resource availability, but traffic volume -- it is both acceptable and desirable to employ a less granular approach to traffic management. In this environment, attempting to maintain and process hundreds of thousands of individual flows, flow states, and resource availability is unrealistic. Another IETF initiative, Differentiated Services (DiffServ), provides many of the traffic management benefits of IntSev without the signaling and state management overhead, and by aggregating large numbers of flows into a few simple Behavior Aggregates (BAs), provides an attractive solution for the core network.

TRAFFIC ENGINEERING
Once appropriately classified and groomed, traffic engineering services must be applied to efficiently aggregate and map service flows onto the existing network topology to control network behavior and optimize network utilization and traffic performance.

MPLS represents the best alternative for enabling traffic engineering and QoS in the heterogeneous public networks. Although originally intended as a means to enhance routing performance, continued improvements in that area have shifted the application focus of MPLS to its inherent capabilities for delivering efficient and scalable traffic engineering and QoS in IP-based networks. MPLS operates at Layer two-and-a-half (L2.5) and is protocol agnostic to the layers above and below it. Its architecture is based upon the multi-layer switch concept, which cleanly separates the forwarding and control functions -- both of which are defined by MPLS.

The power of MPLS stems from its ability to associate and allocate any type of user traffic with a particular Forwarding Equivalency Class (FEC). Each FEC represents an aggregation of traffic that will be treated in the same manner as it traverses the network. These FECs are then mapped to Label Switched Paths (LSPs) that have been engineered to support specific traffic QoS requirements -- guaranteed bandwidth or low latency, for instance. LSPs behave in a similar fashion to the more familiar ATM Virtual Circuit and Frame Relay Data Link Connection Identifier (DLCI), but do so with much greater efficiencies.

Upon ingress to a particular MPLS domain, all packets are assigned a label that serves to represent its FEC/LSP binding as well as a short-hand reference to the contents of the IP packet header. In sharp contrast to today's "longest-match" routing paradigm, MPLS-equipped routers are able to perform ultra-fast forwarding of IP packets via "exact match" label swapping. Moreover, MPLS overcomes the inherent limitations of traditional destination-based routing by supporting both explicit and constraint-based approaches for establishing LSPs. This capability allows network administrators to bypass potential points of congestion and direct traffic away from the default path selected by today's Interior Gateway Protocol (IGP)-based networks and deliver precise control of network traffic and behavior.

PUTTING IT ALL TOGETHER ... IP QOS FOR SERVICE-AWARE NETWORKS
For the resource-constrained local access loop, MPLS is best matched with the IETF's Integrated Services (IntServ) architecture to achieve service-aware networking. This MPLS-IntServ combination provides explicit QoS controls at the LSP-level for delivering enhanced IP-based services. MPLS/IntServ also provides connection-oriented behavior to connectionless access networks.

Since each LSP is apportioned and allocated access bandwidth in accordance with its traffic contract parameters, the given application receives its appropriate level of QoS. For instance, individual LSPs can be created to support the unique requirements of services such as voice over IP (VoIP), VoMPLS, or premium data applications such as e-commerce or VPNs. In the core network, where explicit end-to-end resource reservations are not practical and additional packet processing overhead is not advantageous, pairing MPLS with another IETF initiative, Differentiated Services (DiffServ) provides the required levels of QoS management.

CONCLUSION
Combining IntServ, DiffServ, and MPLS offers a practical "service-aware" QoS architecture that enables service providers to create and deliver a wide range of IP-based voice, unified communications, and data services over a single access network. This approach enables all services -- packet-based voice, tiered data offerings, and secure VPNs -- to dynamically share the same link and provides network operators the tools to optimize resource utilization on their access networks. Finally, an IntServ/DiffServ/MPLS architecture reduces the complexities and costs associated with operating multiple access networks and provides investment protection by integrating with existing legacy infrastructures, while allowing carriers to follow a smooth migration path to converged, IP-dominated networks.

Mark Veil is product manager at Integral Access. The author may be reached at [email protected] 

[ Return To The November 2000 Table Of Contents ]