Abstract
An ongoing challenge in telecommunication is the integration of a variety of services on broadband access platforms at increasing transmission speed. Traditional Internet services like file transfer, email and web browsing, are carried on the same multi service IP platforms with voice, video and television over IP, online gaming, peer-to-peer and grid networking etc. While broadband access is established as a standard equipment for homes, the networking capacities in the access and the backbone are steadily extended to keep pace with higher traffic volumes. Together with the spectrum of services, the traffic mix on the aggregation levels becomes increasingly versatile with different demands for end-to-end transport in terms of throughput, loss and delay sensitivity. The chapter focuses on planning and traffic engineering for link bandwidth and buffers as main resources in communication platforms based on measurement and statistical properties of traffic growth and variability. We summarize quality of service demands of main Internet applications and mechanisms to control and stabilize the performance of ISP network platforms on different time scales. Load thresholds for link dimensioning are derived with regard to quality of service (QoS) demands and the variability in source and aggregated profiles. Finally, link level and network wide traffic engineering is addressed together with load balancing techniques.
TopIntroduction: Growth And Composition Of Internet Traffic
Information on developing Internet traffic and services does not seem to be available as a global picture. A few periodical reports are provided by standardization bodies and fora, e.g. [5] or in official statistics of countries [1] [16] and some spontaneous sources can be found in market research reports as well as publications from equipment vendors [4] and research institutes [3][15]. The Minnesota Internet Traffic Statistics (MINTS) provides an overview on the home page [20] including links to many relevant sources and measurement data from traffic exchanges in the USA. Odlyzko et al. [15] found Internet traffic to roughly double per year since 1990. Figure 1 shows traffic growth factors based on several previously mentioned sources [1][3][16][20] and Deutsche Telekom’s IP platform [17] in the time frame 2001–2007. The steepest curves still follow the trend of doubling per year, whereas moderate cases have annual growth factors in the range 1.5–2. A current white paper of the router manufacturer Cisco [4] gives estimates on the total traffic growth as well as a breakdown into applications. The main conclusions are a forecast of mean annual growth factors of about 1.45 until 2011 mainly driven by several types of video data transfers from peer-to-peer downloads to IP-TV via multicast and video streaming on demand. Within each ISP platform, the product marketing strategy and deployment steps for new transmission and switching technology on the optical and the IP layer have an influence on different growth phases.
Figure 1. Traffic growth observed on the Internet in different continents
Not all factors are clearly predictable and can be planned in advance. When we differentiate the growth due to an increasing subscriber base and the growth caused by in the data rate per subscriber, then the latter is observed to be almost constant on Deutsche Telekoms’ IP platform from 2002 – 2007. Although this may be unexpected on first glance, the figures can be interpreted when looking at the fraction of file sharers, who generate the major portion of the Internet traffic since the millennium. File sharing was a driver of the demand for broadband access especially in the early deployment phase of digital subscriber lines (DSL).
In 2002 probably most of the subscribers were running P2P file sharing programs which dominated the traffic demand and smoothed the traffic curve to an almost constant rate over day and night as visible in Figure 2. Meanwhile the population of DSL subscribers in Germany is well beyond 10 million and most of them are using the Internet for multiple purpose. While broadband access entered the mass market, P2P file sharing traffic continued to increase, but the fraction of extensive file sharers was reducing and the traffic profiles per user shifted to applications with lower demand than file sharing. Thus the mean traffic volume per subscriber did not follow the increase in access speed from less than 1 Mb/s in 2002 up to 16 Mb/s or more nowadays.
Figure 2. Composition of traffic due to TCP/UDP port measurement
In the future, the increase in the broadband access population will be relatively smaller than in the last years, whereas the access bandwidth and the traffic generated per user has a large potential for growth when video streaming and television over IP will become feasible and popular. Therefore measurement of the components in the Internet application mix is important for prognosis and planning processes for network expansion.
Odlyzko et al. [15] already pointed out that the traffic mix is changing with a shift from Web browsing applications to peer-to-peer file sharing as the driver of traffic growth since the millennium. Figure 2 shows measurement over four days on a link in Deutsche Telekom’s Internet platform in 2003 at the time when file sharing was most dominant. The measurement was obtained on transport layer based on standard or usual TCP/UDP port assignment.
Key Terms in this Chapter
Traffic Engineering: Traffic engineering covers all measures to optimize and control traffic flows in a telecommunication network in order to ensure a maximum throughput and a sufficient QoS level. It is a part of the network planning, operation and management process of an ISP. Concepts for dimensioning, admission control, differentiation of services and failure resilience are included which should ensure a well balanced load level for good performance in normal operation and should keep important services available even in relevant failure scenarios.
Quality of Service/Quality of Experience: Quality of Service (QoS) and quality of experience (QoE) measure how much a telecommunication service achieves the expectations of the users. Quality of service introduces performance criteria for data delivery in transmission networks with regard to throughput, delay, delay jitter and availability as main parameters. Quality of experience directly expresses the view of the user and determines the influence of QoS parameters on the user satisfaction. Methods to evaluate user satisfaction on a scale for the mean opinion score (MOS) are partly standardized e.g. for telephony and partly under research e.g. for video
Traffic Flow: A traffic flow carries data that is exchanged between the terminals during a communication service. In addition to unidirectional flows from a sender to a receiver, multicast or broadcast flows distribute data from a source to many destinations via splitting points in switching nodes of the network. Traffic flows on the Internet are transferred as a sequence of IP packets with protocol headers being added to the payload data to facilitate routing and delivery through the network. Internet standardization refers to an IP flow as a collection of successive packets with the same IP addresses and transport layer ports for source and destination as well as the same type of service marked in the packet headers. The TCP protocol initiates and terminates a flow by setting up and closing a connection in a handshake with the receiver, whereas UDP spontaneously starts and stops to send packets and flows.
Traffic Profiles: Traffic profiles in communication networks refer to the mean rate and the variability of the traffic rate over time. Applications on the Internet generate their specific source traffic profiles, from which they may be classified as narrow- or broadband, constant bit rate or variable rate. Telephony and voice over IP differs in the traffic characteristic from video streaming and background data transfers, such that properties of source traffic profiles can be used to classify the corresponding application type. Traffic profiles are also measured for aggregated traffic on transmission links.
Buffers in Communication Networks: Buffers temporary can store data in terminals and switching systems of communication networks. When data is arriving faster than it can be processed then buffering is useful in order to avoid dropping a part of it. On the other hand, buffers in network elements introduce additional delay which limits their efficiency for real time data transfers. Sender buffers help to reduce the variability of source traffic profiles and may transform a variable rate input stream into a constant rate output. In terminals, buffers are necessary to adjust different delays (delay jitter) of received data units for real time service and to hold it available until the play out time as well as for reordering out of sequence packets for transport protocols like TCP.
Statistical Multiplexing: Statistical multiplexing is performed by switching systems in communication networks that merge data packets from multiple input lines and forward them to multiple outputs in a first come first serve or other scheduling discipline. In this way, many data flows can share capacity on a common transmission path. The aggregation of flows in a multiplexer is governed by statistical laws, such that the entire flow usually shows a smoothed rate variability as compared to single flow components.