686 IEEE COMMUNICATIONS LETTERS, VOL. 8, NO. 11, NOVEMBER 2004 A QoS Index for IP Services to Effectively Support Usage-Based Charging Dario Di Sorte, Mauro Femminella, and Gianluca Reali Abstract—The goal of this letter is to present a model to com- pute a quality-of-service (QoS) index to characterize IP services. Then, we show how such a score may be used in a clear and flex- ible way for defining advanced usage-based tariff criteria to charge QoS guaranteed network services to address the dynamics of the expected future telecommunications scenario. Index Terms—Guaranteed IP service, numerical analysis, usage- based tariff. I. INTRODUCTION T ODAY, the most common charging method in the Internet is based on the flat-rate model, i.e., subscribers pay a fee for accessing the network, independent of the actual use of the service. If the Internet is to become a network supporting dif- ferentiated application and transfer services, the deployment of advanced and expensive architectures to efficiently support quality-of-service (QoS), charging, billing, and accounting must be carried out [1]. In addition, the price definition is expected to greatly influence the selection of the network operator. For these reasons, usage-based tariffs must be applied. In fact, the flat-rate model will likely be inefficient from an economic point of view, since it does not enable charges to be made according to the type and to the perceived QoS. A number of pricing models have been proposed so far [2]. In this letter, we start from the tariff model to charge for IP guaranteed services already sketched in [3], which has the fol- lowing characteristics: — it charges the edge-to-edge service of a single admin- istrative domain for a given class of traffic, described by dual leaky bucket (DLB) parameters [4]; — it is on a per-call basis; — it depends on (i) the QoS level of the transfer service offered, (ii) the amount of network resources afforded by the domain for such a service; — it is based on the duration of the connection and the amount of traffic volume exchanged. This tariff is based on the concept of virtual delay [3]. The virtual delay value is a comprehensive, all-inclusive appraisal of the QoS parameters characterizing the edge-to-edge transfer service within a domain. The novelty of this letter is the defini- tion of a law to compute the virtual delay. Then, we show how Manuscript received March 25, 2004. The associate editor coordinating the review of this letter and approving it for publication was Prof. S. Pierre. This work was performed under co-financing by the MIUR within the Project FIRB- PRIMO. The authors are with Dipartimento di Ingegneria Elettronica e dell’In- formazione (DIEI), University of Perugia, 06125 Perugia, Italy (e-mail: disorte@diei.unipg.it; femminella@diei.unipg.it; reali@diei.unipg.it). Digital Object Identifier 10.1109/LCOMM.2004.837661 it depends on the QoS level and on the amount of available net- work resources, and how it affects the tariff. II. COMPUTATION OF THE VIRTUAL DELAY In this section, our goal is to compute the value of the virtual delay that characterizes an edge-to-edge service offered by a domain, from technical considerations at the IP layer, and to show how such a value may be used for charging purpose. We start from the following assumptions. — Flows entering the domain are regulated by DLBs, with parameters: peak rate, , sustainable rate, , and burst tolerance, . These traffic parameters define the extreme traffic profile associated with the consid- ered class of traffic that users are not allowed violating. — The QoS of the port-to-port IP service provided to the specific traffic class is described by the following ser- vice parameters: the maximum transfer delay, , the maximum delay jitter, , and the loss probability, , due to buffer overflow. We assume that these pa- rameters can be negotiated between network operators and users. — Other QoS parameters, such as channel reliability, re- silience and connection set-up time (network parame- ters), characterize the intrinsic quality of the network, do not depend on the specific flow, and cannot be ne- gotiated. — Resource reservation and admission control functions are implemented to respect the service level agreement. Consequently, an amount of network resources is re- served for the flow from the ingress to the egress point of the domain; we are not interested in the QoS archi- tecture, nor in the specific admission control scheme and resource reservation strategy, implemented within the network. The total edge-to-edge delay includes transmission time at the source, propagation delay, processing and transmission time, and queueing delay at network nodes. Since delay jitter is mainly caused by queueing in nodes, it is possible to assume that is a component of the maximum transfer delay. In other words, the maximum queueing delay is equal to the maximum delay jitter. Moreover, in an equivalent “virtual” model, it is possible to compensate the packet loss probability, due to buffer overflow in nodes, by increasing the amount of buffer allocated to the flow. From this point of view, loss probability may be traded for queueing delay, and therefore it may be represented by a contribution in the virtual delay. 1089-7798/04$20.00 © 2004 IEEE