A simple framework for QoS provisioning in traffic engineered networks S. Avallone and G. Ventre COMICS Lab, Dipartimento di Informatica e Sistemistica Universit` a di Napoli Federico II Via Claudio 21, 80125 Napoli, Italy Abstract— Different architectures have been proposed and standardized to support Quality of Service (QoS) in the Internet. The goal was to provide the users with a certain level of QoS guarantees. The Traffic Engineering (TE) concept was then introduced to account for resource optimization, as well as users’ QoS requirements, and inspired new architectures. Routing has also been a flourishing research field. Proposed QoS routing algorithms seek a multi-constrained optimal path, while traffic engineering algorithms aim at maximizing throughput and the number of admitted requests. However, architectures and routing algorithms have been developed rather independently of each other. The aim of this work is to present a framework which provides predictable communication services to flows without requiring sophisticated queuing at individual routers. We show that the routing algorithm plays a key role to assure both QoS and resource optimization. I. I NTRODUCTION In the recent past years several architectures (e.g., IntServ, DiffServ, ATM) have been proposed and standardized with the primary goal of providing different levels of QoS. However, they pay little attention to service providers’ viewpoint. In- ternet Traffic Engineering (TE) is a more general framework that addresses traffic oriented performance requirements, while utilizing network resources economically and reliably. Several works proposed architectures that employ traffic engineering techniques for network management purposes (e.g., [1][2]). Despite TE is a more general concept than QoS, many of such architectures only focused on resource optimization. The research on QoS provisioning and traffic engineering has also produced plenty of routing algorithms. We can distin- guish between QoS routing algorithms and traffic engineering algorithms. The former represent QoS requirements as con- straints and address the problem of finding a multi-constrained optimal path. But, they typically ignore the perspective of service providers. The latter target resource optimization and attempt to maximize the throughput and the number of ac- cepted requests. But, they consider the bandwidth (or the effective bandwidth) as the unique QoS requirement. Typically, architectural considerations and routing algo- rithms are not coupled in previous work. We believe that an architecture aiming at providing QoS cannot prescind from an appropriate routing strategy, as it is difficult to offer strict QoS guarantees without having the control over the route taken by packets. The routing algorithm is the key not only to satisfy the QoS requirements of the flows but also to meet traffic engineering objectives. Also, if properly designed, a routing scheme may avoid the employment of complicated QoS mechanisms in the core of the network, which is the main reason preventing network operators from deploying QoS infrastructures. The approach we propose in this paper succeeds in providing each individual flow with quantitative guarantees by means of as simple scheduling policies as FCFS. This result is achieved through a proper link weight assignment scheme and a constraint-based routing algorithm. II. THE PROBLEM OF QOS LINK WEIGHTS SETTING We model the network topology at time t as a graph G(V,E), V being the set of nodes and E the set of links. Each link l is assigned a QoS link weight vector with as components the available bandwidth w 0 (l; t) ≥ 0 and m additive QoS weights w i (l; t) ≥ 0, i =1,...m. The QoS requirements of a flow are expressed as an (m + 1)-dimensional vector  Q = [Q 0 ,...Q m ], where Q 0 is the requested bandwidth and Q 1 ,...Q m are constraints on the considered additive measures. The goal of a QoS routing algorithm is to find a path P at time t 0 (the arrival time of the flow) such that: w 0 (P ; t 0 ) def = min l∈P w 0 (l; t 0 ) ≥ Q 0 (1) w i (P ; t 0 ) def =  l∈P w i (l; t 0 ) ≤ Q i ∀i =1,...m (2) In order for a QoS routing algorithm to be implemented in practice, it is necessary to determine how to set QoS link weights. This section addresses such a problem, which has not been adequately investigated up to now. We distinguish among bottleneck and additive QoS measures. It is safe to state that the link weight associated with the available bandwidth, the most commonly used bottleneck QoS measure, should be as close as possible to the current bandwidth availability. But, does such a principle hold for additive QoS measures too? An exact QoS routing algorithm seeks a path P such that the inequalities (2) hold. But, the QoS requirements of a flow are satisfied if the QoS experienced across path P is within the constraints for the whole duration of the flow. We denote by q i (l; t) the average value of the i-th QoS measure experienced across link l at time t. Consequently, the QoS requirements of a flow routed along a path P are satisfied if:  l∈P q i (l; t) ≤ Q i ∀i =1,...m ∀t ∈ T (3) 1-4244-0476-2/06/$20.00 ©2006 IEEE. 279