320 IEEE COMMUNICATIONS LETTERS, VOL. 8, NO. 5, MAY2004 Proportional Differentiated Admission Control Ronaldo M. Salles, Student Member, IEEE, and Javier A. Barria, Member, IEEE Abstract—This letter presents a new admission control policy inspired in the framework of proportional differentiated services (PDS). While most of previous PDS has focused on average queueing delays and packet drops to differentiate the perfor- mance of adaptive applications, the proportional differentiation admission control (PDAC) differentiates inelastic traffic in terms of blocking probabilities. The PDAC is built up using asymptotic approximation theory, employs a class based approach, and conforms with the PDS requirements of predictability and control- lability. Numerical experiments confirm a good performance of the approach. Index Terms—Admission control, asymptotic approximation, proportional differentiated services. I. INTRODUCTION A DMISSION control is fundamental for the adequate support of inelastic traffic in data networks. This type of traffic is characterized by stringent quality of service (QoS) requirements and usually do not adapt to network congestion signals (e.g., real-time traffic). Regarding IP networks, the IETF IntServ and DiffServ architectures have emerged as the most prominent proposals to attend the QoS requirements posed by current network applications. However, while the first one may suffer from scalability problems, the latter may only provide qualitative assurances and lacks control to the network operator. Recently, a scalable, predictable (differentiation is independent of class load variations), and controllable (network operator should be able to adjust the relative QoS between classes) service architecture, the proportional differentiated services (PDS) [1], has been proposed as an alternative to overcome those problems. Most PDS implementations do not employ admission control or resource reservation and so they do not serve for inelastic traffic. However a QoS network offering PDS may be also re- quired to transmit inelastic traffic. The proportional differenti- ated admission control (PDAC) bridges that gap and brings the PDS framework to the context of inelastic traffic. It provides a predictable and controllable network service for inelastic traffic in terms of blocking probability, and uses a PDS equality to de- fine such service. Manuscript received September 18, 2003. The associate editor coordinating the review of this letter and approving it for publication was Prof. M. Devetsiki- otis. The work of R. M. Salles was supported by the CNPq, Brazilian Govern- ment, under Grant 200049/99-2. The authors are with the Department of Electrical and Electronic En- gineering, Imperial College London, London SW7 2BT, U.K. (e-mail: r.salles@imperial.ac.uk; j.barria@imperial.ac.uk). Digital Object Identifier 10.1109/LCOMM.2004.827384 II. THE PROPORTIONAL DIFFENTIATION ADMISSION CONTROL (PDAC) A. Environment and Formulation We consider a heterogeneous system where each connection has an associated bandwidth requirement that should be guar- anteed end-to-end throughout its lifetime. To facilitate scala- bility, connections are grouped into classes that differ according to . A connection should use class if , and is the nearest to among all classes. Same class connections are ag- gregated into path flows so that they can be managed as a single entity inside the network The aggregate requirement of class flow is then , where is the number of class connec- tions. A simple admission test can be used to control the access to class partition . This class-based ap- proach is conservative since we assumed traffic to be inelastic and provided in an all-or-nothing basis. Signalling is manda- tory, however, as long as we are conservative (i.e., use as the peak rate) per-flow management is reduced. There will be no need to allocate resources per-flow since class partition is likely to be over provisioned. Moreover, there will be no wasted re- sources given that all available resources can be used by other underlying PDS or best-effort traffic. The above characterization is fundamentally different from the adaptive traffic commonly assumed in PDS. In our case, the only way to regulate the aggregated class flow and still meet bandwidth requirements is through blocking. Note also that blocking is a QoS parameter relevant for inelastic traffic. Each class incurs a certain blocking probability and ac- cording to proportional differentiation goal, PDAC is defined as (1) where, is the weight of class . The weights work as priorities assigned to the classes, the larger the weight the higher the ad- mittance priority. The strategy ensures a flexible traffic control for the network operator (controllability). The problem is then to find a resource allocation mechanism for the link capacity which guarantees (1). The system can be modeled as a stochastic knapsack [2] of capacity , which under a complete partitioning scheme (CP) is reduced to independent Erlang subsystems (exponential arrivals and general session holding times). Each class has an allotted partition and offered load . The PDAC problem under CP is then find (2) s.t. (3) (4) 1089-7798/04$20.00 © 2004 IEEE