Statistical Multiplexing

Twenty-first century communications will be dominated by intelligent high-speed information networks. Ubiquitous access to the network through wired and wireless technology, adaptable network systems, and broadband transmission capacities will enable networks to integrate and transmit media-rich services within specified standards of delivery. This vision has driven the standardization and evolution of broadband integrated services network (B-ISDN) technology since the early 1990s. The narrowband ISDN proposal for integrating voice, data, and video on the telephone line was the precursor to broadband services and asynchronous transport mode (ATM) technology. The ISDN and B-ISDN recommendations are put forth in a series of documents published by the International Telecommunication Union Telecommunication Standardization Sector (ITU-T), which was formerly CCITT [1,2] and the ATM Forum study groups [3,4]. The concept of multiplexing integrated services traffic on a common channel for efficient utilization of the transmission link capacity is central in the design of ISDN and ATM networks. ATM in particular advocates an asynchronous allocation of time slots in a time-division multiplexed frame for servicing the variable-bit-rate (VBR) traffic generated from video and data services. ATM multiplexing relies on the transport of information using fixed-size cells of 53 bytes in length and the application of fast cell-switching architectures made possible by advances in digital technology. It utilizes the concepts of both circuit and packet switching by creating virtual circuits that carry VBR streams generated by multiplexing ATM cells from voice, video, and data sources. The ATM architecture is designed to efficiently transport traffic sources that alternate between bursts of transmission activity and periods of no activity. It also supports traffic sources with continuously changing transmission rates. One measure of traffic burstiness is the ratio of peak to average rate of the source. A circuit-switched network would conservatively allocate to each source a capacity equal to its peak rate. In this case, full resource utilization takes place only when all of the sources transmit at their peak rates. This is typically a low-probability event when the sources are statistically independent of each other. A statistical multiplexer, however, allocates a capacity that lies between the average and peak rates and buffers the traffic during periods when demand exceeds channel capacity. The process of buffering the multiplexed stream smooths the relatively high variations in the traffic rate of individual sources. The multiplexed traffic is expected to have a smaller variance about the mean rate in the limit as the number of sources multiplexed increase to a large value. As a result, there is a diminishing magnitude in the probability of occurrence of source rates that are greater than the available capacity. This feature leads to the economies of scale paradigm of statistical multiplexing that is at the core of B-ISDN and ATM transmission technologies. Statistical multiplexers have been integral components in packet switches and routers on data networks since the 1960s. They have gained increased prominence since 1990 with the availability of broadband transmission speeds exceeding 155 Mbps and ranging upto 10 Gbps in the core of the network. In conjunction with gigabit switching speeds, the new-generation of internetworks have the hardware infrastructure for delivering broadband services. However, since broadband traffic features are highly unpredictable, the control of service quality such as packet delay and loss probabilities must be managed by a suite of intelligent and adaptable protocols. The development of these techniques is the present focus of standards bodies, researchers, and developers in industry. New Internet protocols and services are currently being proposed by the Internet Engineering Task Force (IETF) to enable integrated access and controlled delivery of multimedia services on the existing Internet packet-switching architecture [5,6]. These include integrated (Intserv) and differentiated (Diffserv) services [7], multiprotocol label switching (MPLS) [8], and resource reservation protocols (RSVPs) [9]. It is expected that ATM and the Internet will coexist with ATM infrastructure deployed at the corporate, enterprise, and private network levels. The Internet will continue to serve connectivity on the wide-area network scale. The design and performance of these new protocols and services will depend on the traffic patterns of voice, video, and data sources and their influence on queues in statistical multiplexers. These problems have been the focus of numerous studies since 1990. This article is organized as follows. Section 2 describes stochastic traffic descriptors and models that have been applied to characterize voice, video, and data traffic. In Section 3 the methods applied for performance analysis of queues driven by the aforementioned models are discussed. Section 4 concludes with a discussion on the open problems in this area.