January 2001

BY VIJAY KRISHNAMOORTHY

Any network, regardless of available bandwidth, needs resource management. Much like handling a budget, it needs to be managed well for effective use. Similarly, a network needs solid management to ensure that its most critical applications are given the resources they require. In a broadband access network such as cable or xDSL, bandwidth-hungry applications can often eat into other applications' bandwidth. Even if this situation is controlled, applications such as VoIP can introduce more constraints, strict delay, and delay variation (jitter) requirements.

Quality of Service (QoS) is an essential element of the well-managed network, bringing predictability and availability as a service to applications. QoS is the set of techniques designed to manage bandwidth, delay, jitter, and packet loss in a network. In a pragmatic sense, "QoS is managed unfairness." If infinite resources were available, networks would provide everything to all applications. In the real world, with limited resources and various bottlenecks, it is extremely important to perform proper resource allocation in a network.

Through the resource management enabled by QoS, service providers can offer better service level agreements (SLAs) to customers and generate revenue based on these differential services. For example, a service provider may offer Premium, Gold, and Best-Effort services. For VoIP, Premium could offer assurances on delay and jitter, as well as bandwidth. Gold service may imply assured bandwidth with no tight bounds on delay and jitter, while best-effort service might be thought of as a simple and basic offering. Premium and Gold services are value-added offerings and thus generate revenue. From a customer standpoint, these differential services provide assurances for mission-critical voice, video, and data.

QoS can be provisioned at Layer2, such as DOCSIS (broadband cable) or Ethernet 802.1 Q/p, and Layer3. As an example of a Layer2 technology, ATM offers different classes of service (CoS). Software running on routers and switches provides extensive capabilities to map between Layer2 and Layer3 QoS mechanisms. Since xDSL relies on ATM as the Layer2 transport mechanism, QoS can be achieved using these ATM CoS mechanisms.

As another broadband access network, cable has much to offer in the QoS arena. The DOCSIS v1.1 specification allows for Layer2 QoS signaling and differential services. Different levels of QoS signaling exist between the components' basic cable architecture. For example, the cable modem (CM) and cable modem Termination Systems (CMTS) can signal dynamic flows, such as Telnet or http, between them.

Layer2 QoS is just the beginning. For scalable end-to-end QoS, Layer3 IP mechanisms are also essential.

For IP QoS, the IETF models of IntServ and DiffServ are applicable to all broadband networks. Through the IntServ model, per-flow resource requirements can be signaled using resource reservation protocol (RSVP), which is also an integral part of the QoS architecture. If the reservation is admitted, the DOCSIS Layer can ensure availability of the necessary resources between the CM and the CMTS.

If per-flow QoS is not desired, coarse-grained IP-CoS can be applied in a network. Under the IETF DiffServ model, traffic is divided into multiple classes defined by the network, based on the user needs. Bandwidth, delay, and jitter assurances are created for these classes as opposed to each individual flow within a class. This model applies to almost any network at Layer3. The model classifies packets based on various criteria such as Layer3 addresses and Layer4 ports and marks them with the DiffServ CodePoint (DSCP) in the type of service (TOS) byte, applying the appropriate Per-Hop Behavior to these classes. Traffic may also be conditioned at the edge of a network so that it complies with predefined or agreed upon rates.

In a real-world network, signaling, provisioning, billing, and management are all different aspects of QoS. By focusing on an end-to-end solution to application needs, a QoS-enabled network offers important benefits to both service providers and MSOs and their customers: It allows service providers to generate additional revenue while making the network more predictable and efficient for their clients.

Even in the future, when bandwidth will be plentiful, QoS will remain an essential tool, not only in provisioning value-added services, but also in ensuring "predictability" in a network. No matter how large the available bandwidth may be, a network still needs a strong QoS management model to deal effectively with the congestion points and bottlenecks that inevitably occur due to speed-mismatch or diverse traffic patterns.

Future trends to watch for will include simplification and automation of the QoS models, seamless end-to-end assurances, dynamic provisioning from a service provider and MSO perspective, and a lowered cost of assured QoS services. 

Vijay Krishnamoorthy is a product manager for QoS in Cisco's IOS Technologies Division (ITD), specializing in Service Provider networks. He actively participates in the IETF DiffServ and MPLS Working Groups and has a background in Computer Science. 

[ Return To The January 2001 Table Of Contents ]