Int. J. Electron. Commun. (AEÜ) 60 (2006) 40 – 44 www.elsevier.de/aeue Optimal IP network dimensioning for stream-type traffic using probabilistic delay constraints Sanaa Sharafeddine a , ∗ , Anton Riedl b , Thomas Bauschert c , Dominic A. Schupke c , Jochen Frings c a Lebanese American University (LAU), Beirut, Lebanon b Christopher Newport University, VA, USA c Siemens, Munich, Germany Dedicated to Professor Jörg Eberspächer on the occasion of his 60th birthday Abstract In this paper, we address the problem of capacity assignment (CA) for latency-sensitive traffic in multi-service IP networks. The CA problem is formulated as an optimization problem with nonlinear constraints, where the total link capacity shares allocated to the given service are minimized subject to performance constraints in terms of end-to-end delay requirements. We introduce a new extension to traditional CA problems by setting a statistical bound on the maximum end-to-end delay experienced by all packets associated with the given latency-sensitive service. For a sample IP network scenario withVoIP traffic, we show that a relatively slight increase in capacity as compared with the classical approach is required to keep the probability of exceeding the maximum end-to-end network delay threshold below 0.1%.  2005 Elsevier GmbH. All rights reserved. Keywords: Capacity assignment; Quality of service; Latency-sensitive traffic in IP networks 1. Introduction Latency-sensitive services required by real-time commu- nications are migrating to IP networks with the expecta- tion of significant cost-savings, simplified administration and new revenue sources. One crucial factor for the per- ceived quality of the emerging latency-sensitive services in IP networks is the end-to-end network delay of traffic sam- ples between the sender and the receiver. The one-way net- work delay value should be constrained below a certain threshold to enable interactive communications. In order to guarantee a deterministic delay threshold for all connections at any time, an unreasonable amount of ca- pacity is required leading to very low network utilization and, thus, waste of resources [1]. Based on this fact, we ∗ Corresponding author. E-mail address: sanaa.sharafeddine@lau.edu.lb (S. Sharafeddine). 1434-8411/$ - see front matter  2005 Elsevier GmbH. All rights reserved. doi:10.1016/j.aeue.2005.10.009 propose to use a probabilistic bound for the maximum net- work delay that can occur from any ingress node to any egress node [2]. This helps save a great amount of capacity, while still achieving almost guaranteed quality of service (QoS). We define the maximum delay at one node as the time experienced by the packet that is queued last in the buffer and when the buffer is occupied the most. On an end-to-end basis, the maximum network delay is then the sum of the maximum delays at all nodes along a path. The idea behind considering the maximum delay is: if we are aware of the packets that experience the maximum delay among all pack- ets, we can practically protect all other packets from extra delay and thus from quality degradation. Based on the prob- abilistic network delay constraints, we evaluate the amount of capacity required on each network link while minimizing the total network cost. In packet-switched networks typically supporting connec- tionless non-real-time communications, traditional capacity