Deploying service at the network edge

Now service providers are turning their attention to ensuring IP Quality of Service (QoS) at the network edge. IP networks were originally designed to be best-effort networks, a fine solution for email, chat, Web surfing, and Napster.But as more delay-sensitive and mission-critical applications are being carried across IP backbones, the best-effort delivery system is no longer adequate. Hence, the need emerges for a route to consistent network service delivery over emerging IP networks - IP Quality of Service.

Today's IP backbone is simple in terms of QoS: one size fits all. All traffic is transported with a best-effort service delivery. This architecture has helped IP networks to develop quickly and resiliently. However, the ubiquitous success of IP has prompted the creation of more demanding applications. In addition, many service providers are interested in selling more than just bandwidth. The trend now is to add incremental revenue through new, value-added services.

Surprisingly, an "old" application, voice traffic, is becoming the killer application that is driving the redesign of IP networks. This is primarily because of the application itself. Although certain applications like email see a simple "slow response" rate in times of congestion, real time applications such as voice and video cannot adapt to inconsistent service levels. Delivery delays for voice and video impact the entire application and can make the services completely useless.

The drive to carry voice traffic on IP networks comes from two forces. First, there has been an increase in IP PBXs in an effort to reduce intra-company calling charges. Service providers have the opportunity to connect these sites together with a VoIP transport network. Second, deregulation opportunities have arisen that allow incumbent service providers to expand into new regions. Since there is no established network in the new areas, many providers are planning to build next-generation IP transport networks to carry both voice and data.

Given that IP networks will soon be carrying new delay-sensitive applications, what needs to happen next is the delivery of IP QoS. With the introduction of IP QoS technologies into today's existing network infrastructure, service providers can begin offering IP services that are ready for delay-sensitive traffic. It is important to note that there are QoS responsibilities in all three areas of the network: customer premise, edge and core. At the customer premise the Customer Premise Equipment (CPE) router (or IAD) needs to weigh one type of traffic against another. For example, voice is more important than data. At the edge of the network, the edge router needs to weigh two sets of variables--the type of traffic and the subscriber connections--in order to give the appropriate service. At the core, the infrastructure needs to aggregate all the traffic classes and forward them at the fastest speeds possible

Within each of these areas of responsibility, there are a number of technologies that are emerging. For the customer premise, Type of Service (TOS) is emerging as the IP QoS mechanism of choice. The CPE device examines incoming packets and stamps the packet with a TOS setting based on the IP QoS policy implemented by the IT manager. Typical QoS classes are based on multiple traffic types per subscriber. For example, four traffic classes per 4000 people in a building site equals 16,000 QoS concepts. QoS policies are applied at the subscriber level. At the edge TOS is recognized; queuing is enabled; and the appropriate core-facing technology, either TOS, Multiprotocol Label Switching (MPLS) or Asynchronous Transfer Mode (ATM) is used. The edge router must balance both the traffic classes (voice vs. data) against the subscriber types (business vs. consumer) and classify the packets into the appropriate QoS classes.

Typically, QoS classes are established as follows: four traffic types per subscriber connection times "n" subscriber connections. For example, four classes multiplied by 1,000 sites would result in 4,000 QoS concepts. QoS policies are applied at the business site and traffic class level. At the Core TOS, MPLS, and ATM Virtual Circuits (VCs) are all viable technology choices. Since the core must, as a first priority, perform at maximum rate, the core must primarily deal with packets that are already classified into traffic classes. At the core, QoS classes are typically created as follows: one traffic class per traffic type with a total of two to four QoS concepts. QoS policies are applied at the traffic class level only.

Although all components of the network are important, it is arguable that the edge router exerts the most influence on IP QoS. It is at the edge of the network that packets are classified, prioritized, marked for the rest of the network to understand, and allowed into the network. In order to achieve this packet flow handling, it is imperative that the edge router use hardware technology in order to maintain performance while the processing is going on. Use of ASIC technology is critical for this function.

There are four IP traffic classes that cover the majority of traffic types a service provider may wish to support. Although more traffic classes can be defined, most large service providers agree that four classes is a reasonable place to start:

• The Low-Latency and Low-Jitter (LLLJ) class provides support for real-time, delay-sensitive and jitter-sensitive applications such as VoIP. Packets in this queue are scheduled first to achieve low latency. In addition, the packet queue is shaped to achieve low jitter.

• The Low Latency (LL) class provides that packets in this queue are scheduled first to achieve low latency. This traffic does not require shaping, but it should be rate-limited so that non-low-latency traffic is not completely bandwidth starved.

• The Low Loss class provides that the packets in this traffic class will be given a higher weight in the scheduler. This means more bandwidth will be given and the packets will also be given a higher threshold, or more buffering. This traffic class is often used for mission critical applications where most packets should get into the network but some delays may occur.

• The Best-Effort class means that packets will be scheduled as soon as possible. This class gets the "left-over" bandwidth, but this class should be the least expensive offered to the subscriber.

In order to achieve this packet flow handling, it is imperative that the edge router be built upon an edge-optimized architecture in order to maintain performance while the processing is going on. Use of ASIC technology is critical for this function.

The Best-Effort class means that packets will be scheduled as soon as possible. This class gets the "left-over" bandwidth, but this class should be the least expensive offered to the subscriber.

In summary, service providers are readying their IP network infrastructure to support IP QoS. The technologies are available now and the implementations are underway. In 2002, service providers will begin to offer differentiated IP QoS, not only because this will provide new revenue services, but also because their subscribers will demand it.