Chapter 8: Quality of Service

Overview

This chapter discusses Quality of Service (QoS) and the related subject of traffic engineering. QoS is becoming increasingly important for both private intranets and the Internet and is beginning to have a fundamental affect on the way we design networks, particularly large public networks like the Internet. It is notable that the long-term requirement for QoS mechanisms is a hotly debated topic. It is proposed in some quarters that the increased use of fiber optics, together with Wavelength Division Multiplexing (WDM), will make bandwidth so abundant and inexpensive in the future that QoS will be redundant. Unfortunately, experience tells us that these benefits rarely get passed on to users quickly, if at all, since many service providers still operate virtual monopolies in many regions of the world. We are likely to require more QoS features for the foreseeable future. Although this chapter is concerned with service quality, in reality most IP internetworks today provide only one class of service, called best effort (which essentially means no guarantees for delivery, bandwidth, or response times). In effect, all traffic has the same priority. Over the last decade use of the Internet has increased dramatically for both private and business users. The commercial use of the Internet in particular is a key driver for the implementation of end-to-end service guarantees for e-commerce and business-critical applications; this requires classification and special handling of packet streams at the session or flow level. Businesses are beginning to require tighter Service-Level Agreements (SLAs) from service providers to ensure smooth operations, availability, and guaranteed response times. The SLA comprises a set of parameters that specify such characteristics as bandwidth, buffer use, priority, and CPU use for different types of data.

Another important development is that as bandwidth has increased, so has the ability for application developers to make use of it with new innovative applications. The latest generation of real-time multimedia applications (e.g., interactive videoconferencing and audioconferencing) requires much tighter service guarantees than legacy applications such as FTP, Telnet, and e-mail. Legacy applications are relatively low-bandwidth consumers, fairly insensitive to variable delay (jitter), and do not tolerate packet loss. Real-time applications, on the other hand, are typically high-bandwidth consumers, will often compensate for a degree of packet loss, and are usually sensitive to jitter. Video is especially bandwidth intensive (high-quality video can demand tens of megabytes per seconds), while audio can be transferred over 64-Kbps channels with reasonable quality (CD-quality audio might consume up to 1 Mbps). With both the increase in legacy traffic and the emergence of a new generation of real-time applications, there is now a serious requirement to resolve contention intelligently, by providing defined assurances from the network. Finally, a relatively new application, Virtual Private Networks (VPNs), means that both business and private users will no longer make do with best-effort performance. Providing SLAs at the VPN level can be extremely difficult.

QoS makes no sense from the user perspective unless it is end to end, and in order to provide end-to-end service assurances to meet various application scenarios it is likely that several service classes will be required. One service class will provide predictable Internet services for organizations that do business on the Web. These organizations will be willing to pay more in order to ensure that their services are reliable and timely. This service class may be differentiated as a gold, silver, and bronze service, with decreasing quality and cost. Another service class will be geared toward real-time applications such as Internet telephony and videoconferencing, ensuring tight bounds on delay and jitter. It is likely that organizations will be willing to pay a premium for this service. The best-effort service will remain for those customers who need only connectivity. As we will see in this chapter, one of the most significant problems with providing true end-to-end QoS is that most internetworks are likely to comprise many disparate technologies. Mapping QoS requirements between many different LAN and WAN network technologies can be difficult at best and is likely to require additional protocol standards and a new generation of high-speed switching products. To this end we will discuss two broad architectures developed by the IETF for providing end-to-end QoS: differentiated services and integrated services (including the resource reservation protocol). We also discuss routing and switching enhancements necessary for traffic engineering and scalability, specifically constraint-based routing and multiprotocol label switching.