Pascal Anelli

Performance analysis for an IP Differentiated Services network

Research reported here deals with a communication architecture with guaranteed end-to-end quality of service (QoS) in an IPv6 environment providing differentiated services within a single DiffServ domain. The article successively presents the design principles and services of the proposed architecture, their implementation over a national platform, and experimental measurements evaluating the QoS provided at the user level.

Evaluation of the APS protocol for SDH rings reconfiguration

The synchronous digital hierarchy (SDH) architecture is one of the underlying technologies used by asynchronous transfer mode (ATM) networks. The SDH includes various protection mechanisms. One main design issue is probably the reconfiguration process in case of failure. In case of SDH works on optical fibers with a ring network topology, the automatic protection switching (APS) protocol can be used. This paper addresses the problem of the maximum allowed recovery time in the four fibers ring architecture. We analyze the APS protocol and derive upper bounds for the processing time in each node of the network in order to cope with the maximum reconfiguration time of 50 ms, as specified in the standard. We finally analyze the behavior of the system in case of two interleaved failures. A worst case analysis is carried out, showing that a 100-ms reconfiguration time can be guaranteed.

AIMD penalty shaper to enforce assured service for TCP flows

Many studies explored the guaranteed TCP throughput problem in DiffServ networks. Several new marking schemes have been proposed in order to solve this problem. Even if these marking schemes give good results in the case of per-flow conditioning, they need complex measurements. In this paper we propose a conditioning method to reduce these complex measurements and an AIMD Penalty Shaper (APS) which is able to profile a set of TCP flows so as to improve its conformance to a desired target rate. The main novelty of this shaper is that the shaping applies an AIMD penalty delay which depends on the out-profile losses in a DiffServ network. This penalty shaping can be used with any classic conditionner such as a token bucket marker (TBM) or a time sliding window marker. We made an evaluation of the APS on a real testbed and showed that the proposed scheme is easily deployable and allows for a set of TCP flows to achieve its target rate.

Conception, Implementation, and Evaluation of a QoS-Based Architecture for an IP Environment Supporting Differentiated Services

Research reported in this paper deals with the design of a communication architecture with guaranteed end-to-end quality of service (QoS) in an IPv6 environment providing differentiated services within a single Diff-Serv domain. The paper successively presents the design principles of the proposed architecture, the networking platform on which the architecture has been developed and the experimental measurements validating the IP level mechanisms providing the defined services. Results presented here have been obtained as part of the experiments in the national French project @IRS (Integrated Architecture of Networks and Services).

Impact of Gateways Placement on Clustering Algorithms in Wireless Mesh Networks

In wireless mesh networks, designing algorithms that efficiently balance the traffic loads among a given set of network gateways is a challenging problem. Links interfere, transfer capacity is limited, and traffic demands vary overtime. The position of the gateways also affects the overall network performance as a result of its direct impact on the way routers are associated to gateways. In this paper, we investigate the performance of several routers-to-gateways association heuristics in relation with different gateway placement algorithms.We show that if bounds on the number of hops between routers and gateways exist, load-based heuristics perform the best. In general cases however, interference-based approaches provide better load balancing.

Differentiated Services over Shared Media

The growing use of multimedia communication applications with specific bandwidth and real time delivery requirements has created the need for a new Internet in which traditional best effort datagram delivery can coexist with additional enhanced Quality of Service (QoS) transfers. There are many aspects in QoS control. In this article, we address the problem of the support of Expedited Forwarding over shared media. Shared media can be found in broadcast networks operating in packet mode. One problem in this environment is unsteady bandwidth. On these networks, the total bandwidth which is used depends on the offered load. In case of excess load, the total bandwidth decreases when it should be reaching its maximal value. Therefore, it is difficult to manage the bandwidth since it does not remain at the same level. In this article, we propose a distributed algorithm to manage the bandwidth efficiently and which enables QoS for a DiffServ environment.

Analysis of internet latency: the Reunion Island case

Internet connectivity is not fairly distributed around the world, in particular for islands or isolated areas. An example, the internet connection of Reunion Island is mainly based on links to France located about 10,000kms away. This situation generated a particular connection which induced high delays and degraded internet service. Typically, the minimal delay between France and Reunion Island is around 180ms. In this paper, we investigate the performance of the Internet connection by analyzing delay and path properties from and to Reunion Island mapped to continent IPv4 spread. With two experiments, based on 27 local probes and 7,860,000 traces, we propose a correlation analyzing between delay and path properties. One particular finding is that the delay is more dependent of the chosen path as the geographical distance, compared to models in literature.

Transport congestion events detection (TCED): towards decorrelating congestion detection from TCP

TCP(Transmission Control Protocol) uses a loss-based algorithm to estimate whether the network is congested or not. The main difficulty for this algorithm is to distinguish spurious from real network congestion events. Other research studies have proposed to enhance the reliability of this congestion estimation by modifying the internal TCP algorithm. In this paper, we propose an original congestion event algorithm implemented independently of the TCP source code. Basically, we propose a modular architecture to implement a congestion event detection algorithm to cope with the increasing complexity of the TCP code and we use it to understand why some spurious congestion events might not be detected in some complex cases. We show that our proposal is able to increase the reliability of TCP NewReno congestion detection algorithm that might help to the design of detection criterion independent of the TCP code. We find out that solutions based only on RTT (Round-Trip Time) estimation are not accurate enough to cover all existing cases. Furthermore, we evaluate our algorithm with and without network reordering where other inaccuracies, not previously identified, occur.

An Architecture to Increase Performance of Self-Organizing Networks

The unpredictable variations of the characteristics of wireless and ad hoc networks make Quality of Service (QoS) support difficult. Wireless Mesh Networks constitute an alternative to provide QoS in an ad hoc network. This paper proposes a wireless mesh solution that reduces co-channel interference and simplifies both services and resources management. The architecture is based on a topology control or clustering scheme and a channel assignment mechanism. The network is organized hierarchically in order to obtain a seamless stable backbone. The clustering scheme aims to achieve a maximum number of covered nodes and a minimum hop count, while providing a topology minimizing co-channel interference by means of a channel assignment algorithm. Performance evaluation in NS-2 shows that resource management in a cluster can be uncoupled from that of neighboring clusters and that the network achieves improved performance compared to a classical ad hoc network.

Expedited forwarding for WiFi

It is essential to reserve resources in order to provide an acceptable quality of service in networks with real-time communication requirements. However, such reserved resources, e.g. bandwidth, may be unused as a consequence of the variations in the actual resource demands. Since bandwidth is scarce in wireless LANs (WLANs), QoS provisioning may be very expensive. Therefore, we propose a new resource management approach leading to a more efficient usage of the network in which communicating stations or end-users dynamically hand over some of the free resources temporarily to the other communication neighbors. This paper concentrates on two fundamental problems of such a demand-based sharing of resources: the current estimation of resource utilization, and the algorithm to share and redistribute resources with real-time requirements. This approach for resource and traffic management allows one to achieve significantly better utilization of network resources.