A Unified Metric for Quality of Service Quantification Pedro Alipio Critical Software S.A. Parque Industrial de Taveiro 3045-504 Coimbra, Portugal Solange Rito Lima University of Minho Department of Informatics 4710-057 Braga, Portugal Paulo Carvalho * University of Minho Department of Informatics 4710-057 Braga, Portugal ABSTRACT Internet service providers usually express the quality of network services through a set of values determined according to several network performance parameters periodically collected or measured. However, for common end-users, these values do not give an over- all idea of the quality of the network services as they stand for dif- ferent units and evaluate different perspectives of each service qual- ity. In this context, this paper proposes the definition of a service- oriented unified metric which quantifies a global Quality of Service (QoS) indication by processing standard QoS parameters through a fuzzy controller. The proposed methodology, based on fuzzy logic and tested on Xfuzzy 3.0 platform, allows to close the gap between a high-level QoS perspective and the effective QoS measurements at lower protocolar levels. The definition of a single per-service QoS metric can be useful to simplify control tasks such as QoS routing, SLA negotiation and auditing. Categories and Subject Descriptors C.2 [Computer-Communications Networks]: Network Opera- tions; I.2 [Artificial Intelligence]: Deduction and Theorem Prov- ing General Terms Management, Measurement, Performance Keywords Quality of service (QoS), QoS metrics, Fuzzy sets, Fuzzy logic 1. INTRODUCTION The management of today’s multiservice networks strongly re- lies on the assessment and control of each service quality levels. Depending on each service characteristics, the quality of service (QoS) offered to user applications and services is evaluated through a set of specific metrics. The Telecommunication Standardization * Corresponding Author (pmc(at)di.uminho.pt) Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. SIMUTools/QoSim 2009 2-6 March, Rome, Italy Copyright 2009 ICST ISBN 978-963-9799-45-5. Sector of the International Telecommunication Union (ITU-T) and the IP Performance Metrics (IPPM) working group of the Internet Engineering Task Force (IETF) have been committed on defining concrete metrics for measuring the quality, performance, and relia- bility of Internet delivery services [3, 4, 8]. The defined set of QoS and performance parameters, although very useful from a traffic en- gineering point-of-view, are far from the common user perception and understanding, making difficult service negotiation and audit- ing. In fact, when establishing a Service Level Agreement (SLA) and corresponding Service Level Specifications (SLSs), QoS re- quirements are frequently expressed by less obvious parameters such as: (i) a delay expressed either as the worst case bound (e.g. delay is less then 100 ms) or as a quantile (e.g. delay is less then 20 ms for 98% of packets during 5 minutes); (ii) a delay variation (jitter) expressed either as the bound or as the quantile; and (iii) a packet loss ratio. SLSs may also include qualitative performance parameters instead or in addition to quantitative parameters. An ex- ample of a qualitative parameter is the delay expressed through the linguistic values low, medium or high. The semantics of these pa- rameters and how they are mapped to specific values (or interval) is mainly derived from the QoS definition at the lower level, i.e. they may differ depending on the network infrastructure (e.g. Diffserv, MPLS or ATM). In this scenario, the present study proposes a novel and simple strategy to derive high-level unified QoS metrics for Internet ser- vices resorting to fuzzy logic principles [9]. Attending to the speci- ficity of the problem, which combines the difficulty of handling multiple low-level QoS parameters with the blur boundaries of user perceived QoS, the use of fuzzy logic to achieve a unique per ser- vice QoS metric brings a clear advantage and simplicity to the solu- tion. Fuzzy logic is conceptually easy to understand, it is tolerant to imprecise data, it can model sets through non-linear functions of ar- bitrary complexity and the rules are written using natural language. Thus, it is in fact very suitable to solve these type of problems as it allows mapping measurements into fuzzy sets describing each of the parameter’s values and it includes the proper inference mecha- nisms to reason over rules describing the service requirements re- sulting on a unified metric expressing the overall quality of each Internet service. Although several works have resorted to fuzzy theory principles to model QoS control solutions [5, 2], this paper provides a new contribution in the field of QoS measurement and monitoring by proposing a fuzzy controller for mapping multi-metric QoS de- scriptions into a single value, providing a macro indicator of the service quality. The remaining of this article has the following structure: first, a brief overview of fuzzy logic is provided in Section 2; the fuzzy controller for generating a unified QoS metric is specified in Sec-