Internet tends progressively to be more interactive by supporting real-time communications into this packet-based environment. It is anticipated that online applications will rapidly develop and contribute by a significant amount of traffic in the near future. These applications have strict requirements in terms of delay and throughput which can't be met only if they are prior. However, at high demand of real-time traffic, elastic traffic (which is transported generally by TCP) may not have sufficient resources to be transported with reasonable quality of service (QoS). Therefore, it is primordial to manage properly the resource-network in order to provide the QoS required by both each type of services. In this paper, we present a new fluid model to evaluate the performance of elastic traffic under Low Latency Queuing (LLQ) system combining a priority queue for delay-sensitive applications with a number of Class Based Weighed Fair Queues (CBWFQ) for elastic traffic. The originality of our contribution consists on focusing on the average total number of flows passing through the whole system by approximating it as a lossless best effort system with total load equal to the total carried load of the original system (no blocked load). The core of our analysis is based on some approximations proven for balanced fairness allocation, which provides a reasonable framework for estimating bandwidth sharing among elastic traffic for best effort allocations. Results issued from this allocation are then exploited to deduce the performance of CBWFQ system. Detailed packet level simulations are used to verify the effectiveness and the accuracy of our analysis. The approach presented in this paper allows a rapid performance evaluation of elastic or rate-adaptive traffic circulating in the actual networks.

This paper studies a topical capacitated network design problem that arises in the telecommunication industry. In this problem, given point-to-point demand between various pairs of nodes, a minimum cost survivable network must be designed by installing capacitated equipments (routers, line cards, ...) on nodes as well as link facilities on arcs. This realistic problem finds its motivation in the rapidly developing field of telecommunication networks and the introduction of fiber-optic technology. It is, in particular, faced by the network designer whenever a new network is set up. To the best of our knowledge, this is the first work that addresses this capacitated network design problem where equipments, such as routers and lines cards, have to be settled on network nodes. We present a Tabu Search heuristic to find a solution that minimizes the total network cost. Numerical results are provided for randomly generated networks and networks coming from real-word applications.

Session Initiation Protocol (SIP) is a new signaling protocol used to establish and release sessions between end users. Using SIP over MPLS networks offers new possibilities to enhance network performance and QoS (Quality of Service) perceived by users. In this paper we evaluate a session level approach to enhance QoS in DiffServ domain. We implement scheduling mechanisms for TCP sessions based on mean rate, size and duration. Dynamic declassing of TCP sessions and bandwidth sharing with WFQ system are discussed and tested. These mechanisms are implemented in the SIP Proxy server of the SIP architecture.

The study of the performance of parallel applications may have different reasons. One of them is the planning of the execution architecture to guaranty a quality of service. In this case, it is also necessary to study the computer hardware, the operating system, the network and the users behavior. This paper proposes different models for those based on simulation. First results of this work show the impact of the network and the necessary precision of communication model in fine-grained parallel applications.

The problem studied here concerns the modeling of call blocking in telephone networks. From the usual assumptions such as exponential arrivals and holding time, lost call cleared, the state of the network is described by a finite Markov chain. From the transition probabilities of this process are derived the differential equations associated with the average occupancy of all trunk groups. These traffic equations are simplified by considering independence of blocking for trunk groups in series. The blocking probabilities are estimated using fictitious offered traffic and the Erlang B formula. Such representation takes into account peaky or smooth traffic characteristics. We develop this one-moment model for routing policies such as load sharing and overflow routing. Performances of the model are given in comparison to the solution of the exact Markov chain model or the results of Monte Carlo simulation. Finally, an application to routing optimization and network dimensioning is treated.