Pricing is probably one of the most efficient means to control congestion in a communication network. It is furthermore mandatory for service differentiation.

Our work this year has focused on pricing schemes without bandwidth reservation . It can be decomposed into two main categories: auctioning for bandwidth, and, in a general way, pricing and scheduling.

The first main category is thus auctioning for bandwidth. Our work in this field has been inspired by the progressive second price (PSP) auctions developed at Columbia University . Nevertheless, this pricing scheme presents some drawbacks, such as a time and efficiency-consuming convergence phase and a signalling overhead due to broadcasting the bid profile after each change of bid. These drawbacks have been tackled in , the costly iterative scheme has been replaced by a one-shot multiple-bid scheme keeping the optimality and incentive compatibility properties. This scheme has been adapted in to multicast flows competing with private flows and to the case of an access network through a tree topology . Another adaptation to the case where the utility of getting a given bandwidth depends on the history of allocations (fitting the framework of multimedia for instance) has also been obtained in .

Multi-class pricing is also an important research topic of our group . Following a previous work establishing that, in order to increase its revenue, an ISP should use at its routers a priority scheduling mechanism rather than generalized processor sharing in the case of a simple queuing system, we have extended the result to the case of TCP traffic in competition, since it can modeled at the flow level by a discriminatory processor sharing queue . We have thus focused on priority pricing in , where Aumann-Shaphey prices (that fairly share the total perceived cost) optimizing the social welfare are determined. Similarly, we have studied the pricing of CDMA (wireless) networks using multiple classes ; optimal (in terms of ISP revenue) static prices are determined. In , we have designed a pricing scheme for a RED buffer such that the drop probability (or more exactly the slope of the drop curve of RED) depends on the willingness to pay of the users: the more you pay, the less one of your packets is likely to be dropped. The problem is modelled as a non-cooperative game and conditions for an equilibrium to exist and to be uniquely defined are established.

As other works on pricing, in , we have studied a simple and promising pricing method called the ``Cumulus Pricing Scheme'', dealing with service differentiation and scalability issues. We have mathematically determined parameters optimizing the provider's revenue under the constraint that each user has an incentive to reveal his anticipated bandwidth consumption. In , we have developed in reliability-dependent pricing scheme assuming that the network is uncongested thanks to optic fiber; end-to-end reliability is then the property of interest. In , , a related problem, yield management for computing infrastructures, has been studied in collaboration with IBM, Yorktown Heights. A framework for modelling and analysing the benefits in a scenario of complex dependence of sojourn times of jobs on factors like customers' choice has been considered.

We maintain a research activity in different areas related to dependability, performability, vulnerability analysis of communication systems. In 2005 our effort has been on evaluation techniques using both the Monte Carlo and the Quasi-Monte Carlo approaches. For what concerns the analytical evaluation of dependability measures, we obtained new results on the computation of the moments of such measures by using the stationarity detection. In , we study a passive r/ nredundancy including failures at activation instants ( i. e.,redundant elements only operate in case of failure of working elements and have a non negligible probability of failing at the instant of activations). Under Markovian assumptions, we obtain recurrent and numerically stable expressions of the steady state unavailability, of the mtbfand of the expectation of the first up-time.

Monte Carlo methods represent the single tool to solve very large Markov chains. When dealing with rare events, as in the case of dependability evaluation, acceleration techniques, such as importance sampling, have to be used. In a previous work , we have defined the so-called Bounded Normal Approximation (BNA) property, stating that the normal approximation, and thus the confidence interval coverage is kept bounded as rarity increases. In , , we discuss and relate all robustness measures of rare event estimators: Bounded Relative Error (BRE), BNA and Asymptotic Optimality (AO). We also illustrate on a static reliability estimation problem that they might of the two important components of efficiency: the computational time. We thus define the so-called Bounded Relative Efficiency and Generalized BNA properties which encompass this component. A practical analysis of the coverage is also provided. In we describe how to efficiently compute the sensitivities of the basic network reliability metric in a static context using Monte Carlo, when the failure of the system is a rare event.

In Quasi-Monte Carlo (QMC), the error when estimating an integral uses a deterministic sequence (instead of a random one) called a low discrepancy sequence and having the property to spread quickly over the integration domain. The estimation error is bounded by the product of a quantity depending on the discrepancy of the sequence and the variation of the integrand. But this bound is proved to be useless in practice.

By combining MC and QMC methods, we can benefit from the advantages of both methods: error estimation from MC and convergence speed from QMC. In , we have continued this work. We have especially the validity of the coverage of the resulting confidence intervals. In , it is also shown how, by using QMC (or its randomized version) for a small amount of coordinates, and MC for the remaining ones, the accuracy of the simulation can be improved. The gain is theoretically proved, as well as illustrated on finance problems.

As another remark, due to the correlation structure of the sequence, necessary to ``minimize'' the error, the direct application of QMC methods to the analysis of Markov chains was inefficient. In , we had developed a QMC method analyzing the transient behaviour of discrete time Markov chains using only a two-dimensional low discrepancy sequence. The main characteristic of this method is a relabelling of the chains at each time step of the simulation. In , , the method is adapted to randomized QMC methods and to the case of Markov chains with random horizon (so that regenerative can also be considered). The superiority of the method with respect to those in the literature is again illustrated, as well as proved in some situations. The method is also shown in to improve the simulation of rare events, when combined with importance splitting techniques.

As a consequence of last year's collaboration with Maestro project-team through the TCP and PRIXNET ARCs, we have analyzed the behaviour of two competing TCP connections sharing a common bottleneck link. We have analyzed several loss strategies: fixed loss probability, largest throughput loss or proportional loss. After some derivations in the general asymmetric case, we especially show that in the symmetric, surprisingly, the loss strategy has no consequence on the average throughput. We show, in contrast, that the second moment of the throughput does depend on the strategy . This invariance result has been extended in to more than two connections in competition. This is shown to be of practical importance as it elucidates the throughput of parallel TCP sockets, an approach used widely to improve throughput performance of bulk data transfers (e.g. gridFTP). The validity of the result has also been illustrated through ns simulations.

This research work is part of the activities of the associated team PAIR (see ).

PSQA is a methodology which allows to quantify the quality of a video or audio (or multimedia) stream at the receiving end, after said stream has passed through a packet network such as the Internet. This quantification is done automatically, and in real time if necessary. The specificity of the approach is that the evaluation obtained is very close to the evaluation that could be done by human observers. The reason is that we use a specific statistical learning tool to capture how humans react (from the perceived quality point of view) when receiving these flows. The tool learns and then behaves similarly to real observers. The statistical tool used to this end is the Random Neural Network model, which is an open queuing network with positive and negative customers. The basic initial papers are for video flows and for audio communications.

To the best of our knowledge, PSQA is the only solution to the problem of assessing the quality of a multimedia stream after travelling through a packet network with the given properties: automatic, real-time if necessary or useful, and accurate (close to the evaluation performed by real human observers). In addition to being able to provide real–time assessments, this approach does very well compared to other metrics proposed in the literature providing better correlation with subjective assessments. This year, we explored its application in a wireless context in two papers: and . The former discusses general issues related to the perceived quality in such an environment. The latter presents material associated with control issues. In , a general presentation of PSQA is provided, with new elements concerning sensitivity computations. This is a consequence of the nice mathematical properties of the RNN tool. See also for a global presentation of PSQA in the case of audio signals.

This research work is part of the activities of the associated team PAIR, and is being jointly done with M. Héctor Cancela (PAIR Uruguayan project leader), from the University of the Republic of Uruguay, in Montevideo.

This team has been put in place in 2000, and its main objectives were the development of techniques and tools for Wide Area Network (WAN) topology design and the development of a way of quantifying perceived quality. For the latter topic, see  .

The problem of WAN topology design is usually decomposed into two main parts, the design of the backbone network (usually with a meshed topology) and the design of the access network (usually with a tree-like or forest-like topology); the objectives usually taken into account is the minimization of the cost of the network topology, but we also consider other performance, dependability, or more in general user perceived quality metrics in this design process. This year we finished a doctoral work on the topic , and the balance of the work is as follows.

In the first two years, there was an important effort in the direction of studying and developing effective metaheuristics for this kind of network design problems. In particular, after exploring other alternatives (see , , for example), we selected GRASP, the Greedy Randomized Adaptive Search Procedure heuristic, as our main optimization technique. This method gives very good results (although it does not guarantee optimality), and works as a template, within which other heuristic ideas can be incorporated. During 2003 we employed this method for the design of a backbone with reliability (connectivity) restrictions, our first results in this area were published in , . In 2004 we continued exploring the potential of the GRASP technique, obtaining results on a more complete test suite for the backbone problem , and tackling the access network design problem. For this last problem we used different heuristics as building blocks for the GRASP method, among them the Random Neural Networks that are also being employed with good success in our technology PSQA (see paragraph ). The method was tested on a large test suite with very good results, which were the object of two publications , .

This year we published in a technique allowing the design of a pseudo-optimal topology for the access network. We used GRASP as the general approach and different local improvement techniques such as RNN. See also for a global view of the work developed in the area during the last 3 years.

If most of the QoS-related problems in the core network can be solved by overprovisioning, congestion may continue to arise in some specific cases: when the bandwidth is either physically limited, like in UMTS networks, or if traffic rerouting due to link failure may lead to an overload. Specific research themes and contributions of ARMOR in this context are the following:

We studied in a model for the NAP protocol, dedicated to the auto-configuration of IPv6 routers. If the auto-configuration of hosts is defined by IPv6 and mandatory, IPv6 routers still have to be manually configured. In order to succeed in new networking domains, a full auto-configuration feature must be offered. NAP offers a fully distributed solution that uses a link state OSPFv3-like approach to perform prefix collision detection and avoidance. In this report, we present a model for NAP and analyze the average and maximum autoconfiguration delay as a function of the network size and the prefix space size.

We study communication problems, QoS management, adaptation of flows and energy used in wireless (mainly ad hoc) networks in cooperation with the laboratory CRISTAL of ENSI, Tunisia. The IEEE 802.11 standard proposes an efficient Power Save Mode (PSM) to spare battery energy in wireless network component. One of the drawbacks of the PSM manifests in the latency implicated by missing nodes in the case of routing. In order to decrease the latency, we proposed a model for the energy conservation, based on synchronized alternations of ``on'' and ``off'' states of network activities of the ad hoc network components .

Despite IEEE 802.11 is now widely deployed over the world, this wireless LAN technology has drawbacks that prevent multimedia applications to work properly. Using simulation study we shown (in collaboration with Thomson) in the 2006-paper that IEEE 802.11 networks have strong QoS issues in triple play home networks even using IEEE 802.11e (WMM), which is the standardisation of a QoS differentiation mechanism for WiFi networks.

Multimedia services are a major application for IEEE 802.11 WLANs and they often use multicast to reach several users on the same WLAN without consuming to much resources. The current standard does not utilize rate adaptation technique for multicasting and this prevent applications using multicast to take advantage of the broadband (up to 54 Mb) capacity of the current WLAN technologies since they have to fall back on a lower rate (often 1 or 2 Mb). We propose a new rate adaptation mechanism (SARM: SNR-based auto rate for multicast) for multicasting multimedia content in IEEE 802.11 WLAN environments. It will appear in  .

In order to propose improvement for the next generation wireless LAN, we worked in collaboration with Seoul National University on a wireless packet scheduling algorithm that uses the multi-state (multi-rate) wireless channel model and performs the packet scheduling by taking into account the channel usage time of each flow. We also studied a new contention avoidance mechanism using ``superslot'' and ``pseudo-collision'' concepts .

The norm IEEE 802.11i aims the authentication and the key distribution in wireless networks. We proposed two new secure mechanisms for fast re-authentication in handover operations , . This work will be extended to propose a new solution for inter-technologies (WiFi-WiMAX) handover management in an operated mobile infrastructure.

Nowadays, reaching services on the Internet highly depends on the use of the DNS (Domain Name System) infrastructure, and DNS security has become strictly necessary. Moreover, such a secure infrastructure (DNSSEC) could be used to secure other applications. The goals of our project are the study and deployment of secure DNS transactions, the development of mechanisms allowing the use of the DNSSEC infrastructure to distribute keys and certificates for secure signalization of Mobile IPv6 and use of IPsec (HIP and/or opportunistic IPsec).

Last year we have analyzed the DNS threats, studied the completeness of DNSSEC and IPsec and specified the requirements for key rollover for DNSSEC  . A France-wide DNSSEC platform has been deployed, it is connected to the international DNSSEC, which shadows the standard DNS. We are particularly interested by the automation of key rollover process and the key management process. The DNSSEC validation process is based on the establishment of a chain of trust between zones. This chain needs a starting point called a trusted key. During the deployment phase, secure zones and unsecure zones could coexist simultaneously. Every resolver has to get a copy of the trusted key for every island of security. Currently we study the trusted keys' rollover problem and two algorithms to solve it. The first one is resistant to key-compromise and places the choice of security level in the resolver's administrator side, which is a natural way to protect a resolver  . The second one enables an automatic and smooth exchange of secure keys when an unsecure zone becomes secure and reduces considerably the number of trusted keys needed in a resolver  , .

The illustration and validation of our concepts, with DNSSEC patches and tools, can be found at http://www.idsa.prd.fr. A global view of the results in this area, over the last 3 years, can be read in .

We have developed an algorithm to compute the sojourn time distribution in the processor sharing, single server queue with Poisson arrivals and phase type distributed service times. In a first step, we established the differential system governing the conditional sojourn times probability distributions in this queue, given the number of customers in the different phases of the PH distribution at the arrival instant of a customer. This differential system is then solved by using a uniformization procedure and an exponential of a matrix. The proposed algorithm precisely consists of computing this exponential with a controlled accuracy. This algorithm is then used in practical cases to investigate the impact of the variability of service times on sojourn times and the validity of the so-called reduced service rate (RSR) approximation, when service times in the different phases are highly dissymmetrical. For two-stage PH distributions, we give conjectures on the limiting behaviour in terms of an M/ M/1PS queue and provide numerical examples to illustrate the different approximations. These results have been published in .

We also obtained new results on the transient analysis of averaged queue length in . This work consists in the analysis of an exponentially averaged queue length over a finite interval of time. Such studies are useful in the context of active queue management where packets are dropped according not only to the current state of the queue but also by considered its past states.

In we have studied a queue with two classes of customers, a single server and FIFO discipline. Class- i( i= 1, 2) customers arrive according to Poisson processes; there is no priority between the two classes. The service of class-1 customers is exponential whereas class-2 customers are served instantaneously. We have obtained exact formulas of geometric form for the stationary probability of each ordered state and for the steady-state distributions of customer numbers in the system.

Using the concept of duality between stochastic processes as defined in ``Continuous-Time Markov Chains'' (W.J. Anderson, 1991, Springer-Verlag), we developed a new approach to obtain closed-form expressions of transient distribution for basic Markovian queues. The idea is to go to discrete time through uniformization, then to use duality to map the problem into transient analysis of absorbing models and then, to use combinatorial techniques to analyze these absorbing chains. See for recent results on this approach.

In we published a chapter introducing fluid models, devoted to the practitioners in the field of telecommunications.

On the basis of the simple formulae and precise algorithms that we had obtained for the distribution of the buffer content of a finite or infinite capacity fluid queue  fed by a Markovian queue or by an M/M/1 queue  , we have obtained in an algorithm to compute the transient distribution of a fluid queue driven by a birth and death process whose birth and death rates are suggested by a chain sequence, for which the stationary probabilities do not exist.

The transient analysis of a fluid queue driven by an M/M/1 queue has been considered in and the general case of the birth and death process has been studied in by means of results from spectral theory.

Any real formal framework does not exist for interoperability testing area, contrary to conformance testing. Our purpose is to provide a formal framework (methods, algorithms and tools) that helps in generating efficient interoperability test suites. The model used here is an automata-like structure called IOLTS (Input Output Labelled Transition Systems). It is an LTS which distinguishes inputs, outputs and internal actions.

We have proposed a formal definition of the notion of interoperability, which has been considered by the testing community as one of the first real contributions in this area  . This year we have studied the problem of quiescence management in interoperability testing. We prove that by taking into account potential quiescence of implementations, we can improve interoperability testing   , . Based on this results we developed algorithms that generate more accurate interoperability tests  , .We also study how a distributed approach (including remote approach) can help in efficiently testing components  .

We have also developed methods and algorithms using formal methods for automatic conformance and interoperability tests generation for mobility protocols. The Mobile IPv6 (MIPv6) have been used to validate these algorithms  , .

We have proposed some solutions for some issues regarding the last step of testing activity, where abstract test suites are translated to obtain executable test suites that are executed against real implementations. The new standardised TTCN3 language and the RIPng protocol have been used for this purpose  , .

On a pragmatic side, we try to validate our solutions for new generation network, mainly IPv6 related protocols. We have generated conformance and interoperability tests for significant IPv6 and 3GPP related protocols, like MIPv6 (Mobile IPv6), IPv4-IPv6 transition mechanisms (NAT-PT, 6to4), RIPng (Routing Internet Protocol for IPv6), etc. These tests have been used for many interoperability events such as the ETSI/Plugtest events (since 2000), and the Japanese TAHI events (since 2001).

We are also involved in the ``IPv6 Ready Logo Programme'' (see http://www.ipv6ready.org) which is a world wide certification programme launched by the IPv6 Forum. We are responsible of the definition of technical requirements for the two phases of this programme. The Phase I started two years ago with real success. The second phase started this year, with stronger requirements.

For three years now, the ARMOR team has launched several activities having the common key work: mobility. Some of these activities belong to other subgroups such as the TEST subgroup. Two mains theme set up the core of this new subgroup: Mobile Access Networks, Ubiquitous Terminals and Networks in Motion.

Operators are interested in the adaptation of existing micro mobility solutions to their operational constraints. Hence, we have participated, in collaboration with France Telecom R&D, to the development of a new micro mobility protocol derived from HMIPv6. This protocol allows the network to take in charge the handover decision. Then the network can prepare the handover to reduce the latency time and perform load-sharing taking into account the overall situation of the network. The new protocol (NCHMIPv6: Network Controlled HMIPv6) is being improved to provide quality of service management in access network.

We are also involved in the definition of a new architecture that allows the applications to be able to adapt efficiently and timely their behaviours to network condition changes. This leads us to deal with automatic interface choice and automatic configuration, as well as with the relation between the network and terminals in term of security and quality of service (on-going work). We have also defined a framework to build adaptive application and to manage network resources description thru a profile mechanism. This work has been recently extended to address the multi-interfaces management in the case of a mobile router . Note that, NEMO (Network in Motion) and its application to Intelligent Transportation System is becoming one of our important topics.

Our work focuses on two main technologies that seem to be contradictory: optical and wireless networks. However, to paraphrase from , we can think of optical fiber and wireless communications as quite complementary. Optical fiber doent go everywhere, but where it doe go, it provides a huge amount of available bandwidth (well over tens of terabits per second over a single fiber). Wireless, on the other hand, does go almost everywhere, but provides a highly bandwidth-constrained transmission channel, susceptible to a variety of impairments.

Optical fiber communication is now ubiquitous in the telecommunications infrastructure. Fiber optics and wavelength-division multiplexing (WDM) technology have increased significantly the transmission capacity of today's transport networks, and they are playing important roles in supporting the rapidly increasing data traffic. Nonetheless, the rigid and large routing granularity (i.e. wavelength) entailed by such an approach could lead to bandwidth waste. In our work, we propose and evaluate a new concept of traffic aggregation in WDM optical networks that aims at eliminating both the bandwidth underutilization and the scalability concerns that are typical of all-optical wavelength-routed networks . Our objective is to reduce the network cost while preserving the benefits of all-optical wavelength-routed networks. In order to assess the efficiency of our proposal, all underlying network costs are compared. Our results show that the proposed aggregation technique can significantly improve the network throughput while reducing its cost .

On the other side, wireless communication is now firmly established as the preferred means of communication for signals over a few megabits per second over distances more than a few hundred meters. Future wireless networks are expected to provide IP-based coverage and efficient mobility support with end-to-end Quality of Service (QoS) requirements. In this work, we propose a new architecture that supports both mobility and QoS management in wireless MPLS networks . To achieve this, we propose a new micro-mobility management scheme called Micro Mobile MPLS . Our proposal includes two protocol variants. In the first variant called FH-Micro Mobile MPLS, we consider the fast handoff mechanism, which anticipates the LSP procedure setup with an adjacent neighbour subnet that an mobile node (MN) is likely to visit. This mechanism is proposed to reduce service disruption by using the link-layer (L2) functionalities. In the second variant called FC-Micro Mobile MPLS, the forwarding chain concept, which is a set of forwarding path, is provided to track efficiently the host mobility within a domain. This concept can significantly reduce the registration updates cost and provide low handoff latency.

Dependability and efficiency in routing and QoS management have to be improved for high speed networks. Our research on this domain is focused on the management of Optical Networks. Some of these works are connected to other works described in the multicast chapter.

Backbone data networks evolve towards all-Optical Networks. In such networks IP routers are interconnected by re-configurable optical cross-connects (OXC) managed by an automatic control plane. This architecture, on the one hand, improves the overall network performances by providing advanced functionalities such as optical by-pass, traffic engineering and fault protection mechanisms but, on the other hand, introduces an additional switching layer which increases the overall network cost. A general presentation of some of the main problems can be found in .

In previous works, we have proposed several heuristic for multicast tree construction. These polynomial time approximations take into account some specific constraints of optical networks: QoS constraint, lightwave constraint and/or constraints on specific resources from optical switches (I.e. limited availability of light splitters, light amplifiers, light swappers, etc.). Some of these works are described in .

Moreover, we have previously proposed to use an enhanced architecture which introduces an intermediate layer based on MPLS . This architecture introduces some independance and flexibility in the network management. This multi-layer architecture have lead us to consider modelling techniques which enable us formulate and to solve various design problems, including multi-layer modelling of networks comprising different technologies and incorporating network recovery mechanisms , .

Dependability is the ability of a network to cope with failures, i.e., to maintain connections established even in case of failures. IP routing protocols (such as OSPF and RIP for instance) do not fit dependability objectives of today applications. Moreover, forwarding technique based on destinatination address (like IP) produce a many to one routing scheme where all paths sharing the same destination must share their protection scheme: protection scheme of paths having the same destination must be established in a coordinated way. In this work , we propose fault recovery for many-to-one routing scheme based on preplanned protection. The main advantage of our approach is that recovery in case of failures is achieved within a short delay. Additionnaly, with respects to other approach, the dependability of the routing scheme is increased because we prove that it statistically withstans more failure types and locations. For each routing scheme our algorithm computes a multi tree where the primary tree could be choosen arbitrarily (for instance the shortest path tree) and the backup tree could be obtain efficiently.

Scalability and efficiency of routing, QoS management and tree construction have to be improved for multicast traffic. For a global description of our contributions, see and . Our research on multicasting can be organized in several thematic axis: