The Role of Admission Control in Assuring Multiple Services Quality Solange Rito Lima, Paulo Carvalho and Vasco Freitas University of Minho, Department of Informatics, 4710-057 Braga, Portugal {solange,pmc,vf}@di.uminho.pt Abstract Considering that network overprovisioning by itself is not always an attainable and everlasting solution, Admis- sion Control (AC) mechanisms are recommended to keep network load controlled and assure the required service quality levels. This article debates the role of AC in mul- tiservice IP networks, providing an overview and discus- sion of current and representative AC approaches, high- lighting their main characteristics, pros and cons regarding the management of network services quality. In this debate, particular emphasis is given to an enhanced monitoring- based AC proposal for assuring multiple service levels in multiclass networks. 1. Introduction To face today’s Internet service heterogeneity and inte- gration, the TCP/IP protocol suite has been enhanced with new service models, protocols and mechanisms. The Class of Service (CoS) paradigm, where flows with similar char- acteristics and service requirements are aggregated in the same class, has been pointed out as a suitable service model regarding scalable Quality of Service (QoS) support. To control network resources efficiently and assure the required QoS levels, Admission Control (AC) has been rec- ognized as a convenient traffic control mechanism [1,2]. In fact, controlling the admission of flows entering the network and sharing a service class aims at avoiding overutilization of existing resources, satisfying the requirements of new in- coming traffic flows without compromising the QoS of al- ready active flows and, generically, preventing instability and congestion assuring QoS and SLSs fulfillment. In general, the QoS guarantees and predictability re- quired by a service class determines the control complex- ity inherent to an AC strategy. To obtain a good compro- mise between service guarantees, complexity and efficient resource utilization is a major challenge. Overprovisioning can be useful to improve this trade-off, however, a consis- tent QoS solution cannot just be based on overprovisioning and further control has to be in place to honor QoS require- ments in the network. The challenge is increased when con- sidering multiservice networks and end-to-end QoS deliv- ery, as service classes have distinct characteristics requiring different QoS assurance levels, and multiple heterogeneous domains may be involved with negotiated SLSs’ between them to be fulfilled. The main objective of this document is to discuss exist- ing AC proposals, covering their main characteristics, ad- vantages and limitations in controlling multiple service lev- els. This analysis is relevant as an effective way to iden- tify, understand and compare representative AC approaches, pointing out strategic directions for improving AC tasks. Facing this discussion, an enhanced AC proposal for man- aging QoS and SLSs in multiclass networks is presented. The remaining of this document is organized as follows: relevant characteristics of existing AC approaches are iden- tified and debated in Section 2; current and representative AC approaches are discussed in Section 3; the characteris- tics and key points of the AC model proposed for QoS and SLS control are highlighted in Section 4. 2. Relevant characteristics of AC approaches Important high-level characteristics distinguishing AC approaches have been identified as follows: (i) the underlying network paradigm - this aspect is re- lated to the network model in which AC operates. AC approaches span from single service (best-effort) to mul- tiservice architectures, following a flow or class-based paradigm. Their scope as regards targeting an intradomain, interdomain and/or end-to-end solution also varies; (ii) the type of service to control - this aspect is closely related to the guarantee levels to be provided. Common and similar terminology includes guaranteed vs. predic- tive, guaranteed vs. controlled load or hard vs. soft real- time services. The type of service is tied up with the appli- cations’ characteristics, whether they are rigid or adaptive, have quantitative or qualitative QoS targets; (iii) the signaling support involved - this topic can be viewed in two distinct ways. On the one hand, it is related to