Department of Electrical Engineering and Computer SciencesUniversity of CaliforniaBerkeley, California 94720In favor of an enhanced network interface for multimedia servicesPaul HaskellCompression Labs, Inc.David G. MesserschmittUniversity of California at BerkeleyABSTRACTThe future will see networks that are increasingly heterogeneous in transport media (such as wireless, public telephone network, Internet, and broadband ATM), terminals (from desktop supercomputers to personal digital assistants with limited wireless bandwidth), applications, and network services (including continuous-media (CM) datatypes in addition to data). Aside from seamlessly supporting this heterogeneity, goals for a well-thought-out network architecture should include high traffic capacity (especially on wireless links, for example through joint source/channel coding), high subjective quality (including low delay), privacy by end-to-end encryption, and enhanced connectivity (including multicast and mobility). Past attempts at defining an internetworking protocol suite for CM services have concentrated on resource reservation for delay guarantees. This paper summarizes a number of other issues relating to CM services, with particular emphasis on wireless access links, pointing out that the highest subjective quality and traffic capacity cannot be obtained by simply adding wireless access to existing broadband networks and compression standards. The simplistic approach of transcoding (conversion from one compression standard to another) has a number of undesirable characteristics, among them a network infrastructure relatively closed to change and inconsistent with privacy. We describe a framework based on decomposing CM (video, audio, graphics, animation) into substreams at the network interface (as proposed in standards like IPv6 and ST-II in the Internet), as well as at internetworking interfaces, and propose other characteristics of this interface. We point to new opportunities tailored to CM services, like annihilation of stale information and progressive delivery services. This research supported by Asahi Semiconductor, Pacific Telesis, Tektronix, MICRO, and the AdvancedProjects Research Agency.Submitted to IEEE Multimedia Magazine. Please send comments to [email protected]. Lastmodified January 3, 1996.21.0 INTRODUCTIONAn “open network architecture”, popularized by the Internet, allows applications and services to be defined by users or third-party vendors transparently across LANs, MANs, and WANs. However, the Internet and its internet protocol (IP) were originally defined for data services, and in their present form do not offer quality of service (QOS) guarantees for continuous-media (CM) services like voice and video, which may be necessary under high traffic conditions. Efforts to extend IP to offer real-time guarantees by appropriate resource reservations are underway [3][4][6][7]. The real-time aspect is but one major issue confronting high-speed networks of the future [1]. Simultaneously, there is underway an effort to define a broad “Global Information Infrastructure” (GII) that subsumes and extends many existing networks [2]. While some existing networks do not interoperate gracefully1, within the context of the future GII one goal is to allow applications and services to span a diverse and heterogeneous set of subnetworks [1][2].Future multimedia networks should support a variety of requirements simultaneously:• Heterogeneity in three dimensions: applications, transport media (including wireless access media), and terminals (ranging from low-capability wireless devices to desktop supercomputers).• Differing connection topologies, including point-to-point, multicast, and multisource. (For example, video conferencing requires both multicast and multisource topologies.)• Untethered (no wires), nomadic (accessible from different locations), and mobile (accessible while moving) multimedia applications will be important for some users.• Privacy by end-to-end encryption will be demanded by a subset of the users, as they gain familiarity with privacy in their data applications.• Traffic efficiency, which is particularly an issue on wireless access links. Wireless access suffers from interference-limited traffic within a given volume of space and bandwidth, and will become an increasingly serious bottleneck as backbone networks become faster.• High subjective quality, which is indirectly impacted by compression algorithms and by loss and corruption in the transport.• Low latency is critical for some interactive applications. Since the propagation delay in terrestrial global networks is significant (hundreds of milliseconds), there is little headroom to increase latency through signal processing and queueing for this subset of interactive applications.One of the greatest implicit challenges is to contain the complexity inherent in provisioning a diverse set ofapplications and services on a heterogeneous infrastructure while meeting all the other objectivessimultaneously.These objectives point to the need for a model for network services that facilitates:• QOS guarantees and QOS selection for real-time services where needed,• High network resource efficiency, particularly on wireless access links,• An architecture that allows new network and signal processing developments to be integrated seamlessly with an existing infrastructure while meeting the other functional and performance objectives.The first requirement argues for a connection-based network model, in which appropriate networkresources are negotiated at the time of service establishment. The second requirement argues for asubstream interface to support joint source/channel coding, as described later. Then, application andnetwork elements can jointly optimize their resource use for a desired service performance level. The third1. For example, how would one place a telephone call through the Internet and the public telephone network?3 requirement argues for careful attention to modularity [15]. Without care, the first two requirements willlead to undesirable dependencies between the implementation of the terminals and transport, leading toinflexibility for the future.We use the terminology architecture of [2] as shown in Figure 1. Applications draw upon the transport services layer, which calls upon the bearer services layer, which carries the bits from one location to another. The transport services