Measurement-Based Admission Control for a Flow-Aware Network Yuming Jiang a , Peder J. Emstad a,b , Anne Nevin a,b , Victor Nicola c , Markus Fidler d a Center for Quantifiable Quality of Service in Communication Systems; b Department of Telematics Norwegian University of Science and Technology, Norway c Department of Electrical Engineering, University of Twente, The Netherlands d Department of Computer Science, RWTH Aachen University, Germany ymjiang@ieee.org, {peder, anne.nevin}@q2s.ntnu.no, nicola@ewi.utwente.nl, fidler@ieee.org Abstract —To provide statistical service guarantee and achieve high network utilization, measurement-based admission control (MBAC) has been studied for over one decade. Many MBAC algorithms have been proposed in the literature. However, most of them belong to aggregate MBAC algorithms which assume or require that (1) First-In-First-Out (FIFO) is used for aggregating flows; (2) statistical service guarantees are provided to the ag- gregate of admitted flows; (3) each flow requires and experiences the same statistical service guarantees as the aggregate. In this paper, we focus on per-flow MBAC that aims to provide possibly different statistical service guarantees to individual flows in an aggregate. Particularly, we propose a simple per-flow MBAC algorithm in which dynamic priority scheduling (DPS) is adopted to aggregate flows. With this DPS-based per-flow MBAC algorithm, a newly admitted flow is always given a lower priority level than all existing flows, and its priority level is improved if an existing flow leaves the system. Consequently, once a flow is admitted, its received service will not be adversely affected by other flows admitted after it. Because of this, there is no need to re-check or adjust network resources allocated to existing flows due to the admission of a new flow. I. I NTRODUCTION While many network applications such as VoIP and stream- ing audio and video are both delay and loss sensitive, they can tolerate some delay and loss. As a result, statistical service guarantees have attracted a lot of research interest in the past decade. For providing statistical service guarantees (SSGs), measurement-based admission control (MBAC) has long been recognized as an important technique because of its ability in achieving high network utilization when ensuring SSGs. In measurement-based admission control, an MBAC al- gorithm uses the a priori source characterizations only for incoming flows and for existing flows that have been in the system, it uses measurements to characterize them. In the literature, many MBAC algorithms have been proposed and investigated [15][6][23]. While these algorithms use different analytical bases for admission test, they commonly assume or require that [15] (1) FIFO is used for aggregating flows; (2) statistical service guarantees are provided to the aggregate of admitted flows; (3) each flow in the aggregate requires and experiences the same statistical service guarantees as the aggregate. We call these algorithms aggregate MBAC algo- rithms. While the above assumptions and requirements have made current MBAC algorithms simple, networked multimedia applications are so diverse that their quality of service (QoS) requirements can be far from each other. In such cases, per- flow MBAC algorithms are preferred [22]. In this paper, we focus on providing SSGs to individual flows. Particularly, we consider a flow-aware network where each flow may have different SSG requirements. We propose an MBAC algorithm that adopts priority to schedule flows in the system and uses measurements of existing traffic to determine if or not a requesting flow can be admitted based on its required SSGs. A newly admitted flow is always given the lowest priority and its priority level is improved if an existing flow leaves the system or other flows are admitted after its admission. For the admission control, both delay and loss are taken into account. Analytical results for the proposed dynamic priority scheduling (DPS) MBAC are presented. The proposed DPS MBAC implies that the experienced SSGs of an admitted flow is not adversely affected by flows admitted after it. The rest is organized as follows. Sec. II introduces flow- aware networking briefly. Sec. III provides some preliminaries for analysis. Sec. IV introduces the DPS MBAC algorithm and its analysis. Sec. V presents some numerical results. Finally, conclusion is made in Sec. VI. II. NETWORK MODEL A. Flow-Aware Networking We consider a flow-aware network. In the network, a flow is defined to be and identified as a set of packets related to an instance of some network application observed at a given network point with an inter-packet interval less than a certain time-out period. Specifically, a flow consists of packets having the same values in certain header fields. A flow is said to have ended or left when no packet with the same header field values is observed for the time-out period. There are several possible ways to identify a flow. One is to use the five-tuple of IP addresses, protocol and port numbers. Another is to use the flow label field in the IP header as specified by IPv6 associated with the source and/or destination addresses. Here, we simply assume each flow can be identified, but how this is done is out of the scope. Flow-aware networking was proposed recently as an alterna- tive QoS architecture for the Internet [5][22]. While an IntServ network also requires flow-level identification, per-flow service