Quality of Service (QoS) has emerged in the last few years as a topical and challenging research area in distributed systems. For a broad definition of QoS we refer to the ISO's Reference Model for Open Distributed Processing (ISO/IEC IS 10746-2): `The notion of QoS is a system or object property, and consists of a set of quality requirements on the collective behaviour of one or more objects . . .'. In the context of this special issue, QoS is primarily associated with systems such as distributed multimedia application platforms or distributed simulations. The QoS requirements of such systems typically relate to measures of rate and latency of information transfer, probability of a communication being disrupted, probability of system failure, probability of storage failure, etc. The role of QoS management is thus to ensure that applications are able to specify and obtain the quality of service that they require for their correct execution.

In general, the functions of QoS management may be subsumed under the following headings: (i) QoS specification and mapping, (ii) QoS negotiation, resource allocation and admission control and (iii) QoS monitoring, adaptation and renegotiation. QoS specification defines the QoS required by an application in terms of both timeliness constraints and guarantees. QoS mapping then translates between QoS representations at different system levels, relieving the user of the necessity of thinking in terms of low-level QoS representations. For example, a user may express a jitter requirement by manipulating a slider in a GUI, and this could be mapped at the lower layers into a requirement for an absolute bound on ATM cell jitter and a jitter smoothing buffer of a certain size.

QoS negotiation, resource allocation and admission control are collectively responsible for configuring a system in preparation for the execution of a QoS requiring application. QoS negotiation provides a framework in which the QoS levels supported by individual system components (e.g. file server, operating system and network) are composed to build the required end-to-end service. Providing such a service may well require explicit resource allocation (e.g. of network bandwidth or CPU cycles); admission control may also be required to ensure that requested QoS levels can be met.

QoS monitoring, adaptation and renegotiation are concerned with supporting QoS during the run time of an application. QoS monitoring and QoS adaptation are conceptually related in a feedback loop in which current QoS levels are observed and fine grained reconfiguration is triggered as necessary to ensure that the required levels are maintained. An example of such a loop is to be found in Internet tools such as vic or vat which monitor network delay and adapt by adjusting the size of an elastic playout buffer. QoS renegotiation is a coarser grained adaptation, often explicitly initiated by the user, in which required QoS levels are revised at run time. An example could be a user scaling a video display from colour to monochrome so that sufficient resources can be freed to support an additional connection.

The papers in this special issue were selected from the fourth QoS Workshop held in Paris in March 1996. To understand the overall evolution of QoS research it is useful to know some background on the series of workshops of which the Paris workshop was a part. The first workshop to be dedicated exclusively to QoS issues was organized in Canada by the European RACE project R2088 (TOPIC) in cooperation with the University of Montreal in June 1994. A European counterpart of this initiating event occurred in the same year in September 1994 in Aachen, Germany. This latter event was held in cooperation with the European Conference on Intelligence in Broadband Services and Networks (IS&N94). A third workshop was then held in conjunction with the IFIP conference on Open Distributed Processing (ICODP95) in Brisbane, Australia in March 1995. By 1996 more independence, and also more recognition of QoS as an independent research area, was achieved in the above mentioned Paris workshop. This workshop was aligned with the IFIP conference on Formal Methods for Open Object-based Distributed Systems (FMOODS96), but was also independently recognised by IFIP. A fifth workshop is due to take place in May 1997 in New York. This event will be the first not to be organized under the umbrella of another big event. This is indicative of the fact that the community interested in QoS issues has now become large enough to maintain its own independent forum of discussion and exchange of expertise.

The five papers in this special issue cover a wide range of QoS research issues and fall broadly into three categories. First we have a paper that addresses a very specific but nevertheless crucial aspect of QoS provision: supporting the QoS of variable bit rate (VBR) traffic in networks with deterministic guarantees. `On the effects of smoothing for deterministic QoS' by Edward Knightly and Paola Rossaro examines the impact of rate smoothing of VBR sources such as MPEG video. The impact of such smoothing on both the QoS as perceived by the user and on resource utilization in the network are considered. Experimental evidence is presented which shows quantitatively how reducing burstiness through source smoothing techniques can significantly reduce the resource utilization of VBR streams while maintaining deterministic QoS for the user (at the cost of some increase in the end-to-end delay bound). It is also shown that a richer traffic characterisation model (D-BIND) can produce better network utilization and user QoS than the usual models based only on peak rate, average rate and burst length.

Next we have two papers concerned with broader issues of QoS management. These papers explore approaches to QoS monitoring and adaptation which are intended to be generic and applicable in a range of implementation environments. `Some principles for quality of service management' by Gregor v Bochmann and Abdelhakim Hafid offers lessons in QoS management learned during the implementation of a prototype News-on-Demand application. Some general principles are extracted from this experience. In particular, a novel QoS adaptation technique is highlighted: transparent automatic reconfiguration of the components involved in a communication (e.g. choice of an alternative network or server at run time). An algorithm which attempts to choose optimal configurations is discussed. `Quality of service management using generic modelling and monitoring techniques', by Leonard Franken and Boudewijn Haverkort investigates the use of Petri nets as the basis of generic QoS monitoring of distributed applications. A distributed application is exploded into finegrained component parts and interactions between these parts are instrumented. The paper offers a case study of the instrumentation of a videophone application using this technique. Simulation is used to evaluate the scheme.

The final two papers in the special issue are more focused and pragmatic in nature. These papers explore QoS provision in particular environments (the World Wide Web and ATM networks respectively) through reported implementation experience. `QoS management in a World Wide Web environment which supports continuous media' by Michael Fry, Aruna Seneviratne, Andreas Vogel and Varuni Witana looks at the practical provision of end to end QoS management in the World Wide Web. The paper looks beyond currently available tools such as RealAudio and StreamWorks and presents a QoS managed RTP based solution featuring an adjunct QoS management protocol. This work offers QoS management functions (e.g. QoS negotiation, adaptation and control of QoS degradation paths) directly to the user via the usual Web GUI. `A QoS adaptive multimedia transport system: design, implementation and experiences' by Andrew Campbell and Geoff Coulson offers further practical experience of QoS management. Their scheme is embedded in an experimental ATM network with the potential for guaranteed QoS. The system features QoS support mechanisms in both the network and the end systems. Of particular interest is reported experience with a dynamic QoS adaptation protocol implemented in the network and based on video scaling techniques and filtering.

In summary, this special issue provides an up to date review of approaches to QoS management and their practical realization. Of course, no claim is made as to comprehensiveness, but the chosen papers do serve as a highly representative sample of current directions in QoS research. The editors are very much obliged to all authors, reviewers and publishers. Without their excellent work, and the contribution of their valuable time this special issue would not have been possible.