INRIA, Evaluation of Theme 1B

1 Administrative aspects

Project-team acronym: ARMOR

Project-team title: ARchitectures et MOdeles de Reseaux(Architectures and models of networks)

Scientific leader: Gerardo Rubino

Research center: IRISA – INRIA Rennes

Common project-team with: CNRS, University of Rennes 1, INSA Rennesand the ENST Bretagne

Personnel (January 2000)

Misc. INRIA CNRS University TotalDR (1) / Professors 1 2 3

CR (2) / Assistant Professors 3 2 1 6Permanent Engineers (3)Temporary Engineers (4) 2 2

PhD Students 10 2 1 13Post-Doc.

Total 13 7 4 24External CollaboratorsVisitors (> 1 month) 2 2

(1) “Senior Research Scientist (Directeur de Recherche)”(2) “Junior Research Scientist (Charge de Recherche)”(3) “Civil servant (CNRS, INRIA, ...)”(4) “Associated with a contract (Ingenieur Expert or Ingenieur Associe)”

Personnel (September 2003)

Misc. INRIA CNRS University TotalDR / Professors 1 3 4

CR / Assistant Professor 5 2 2 9Permanent Engineer 0.5 0.5Temporary Engineer 1 2 1 4

PhD Students 13 4 6 23Post-Doc.

Total 19 9.5 12 40.5External CollaboratorsVisitors (> 1 month)

1

Changes in staff

DR / Professors Misc. INRIA CNRS University totalCR / Assistant Professors

Arrival 3 2 5Leaving 1 1

Current composition of the project-team (sept 2003):

• DR INRIA: Gerardo Rubino (project leader)

• CR INRIA: Bruno Sericola and Bruno Tuffin

• Professors at the University of Rennes 1: Francoise Andre, Bernard Cousin andRaymond Marie

• Assistant Professors at the University of Rennes 1: Louis-Marie Le Ny and CesarViho

• Assistant Professor at INSA Rennes: Miklos Molnar

• Assistant Professors at ENST Bretagne: Jean-Marie Bonnin, Francis Dupont, Lau-rent Toutain and David Ros

• Permanent Engineer: Laurent Guillo (half-time)

• Temporary Engineers: Samir Mohamed, Frederic Roudaut, Bruno Deniaud and OlivierCourtay

• PhD students:

– Sebastien Barbin – QoS and interoperability testing of New Generation IP protocols.

– Ali Boudani – Multicast routing for WANs: routing state reduction for multicastrouters and management algorithms for very large number of small and sparse groups.

– Djalel Chefrour – Cooperation between self-adapting components in mobile environ-ments.

– Joel Corral – Active and passive measures in high-speed networks. Applications toIPv6.

– Thierry Feuzeu-Kwenkeu – Control of versatile, dependable and high speed accessnetworks: auto-configuration, topology and resource discovery processes, rerouting,route optimization, route securisation.

– Sophie Fortin – Performance analysis of transmission and admission control protocols(TCP and AIMD protocols, MAC layer in WLANs of the IEEE 802.11 type).

– Gilles Guette – Enhancement of DNS security: enhancement of DNSsec, integration ofDNSsec and IPv6, 0-configuration tools, enhancement of security in mobile networks.

– Alexandre Guitton – Management of dynamic multicast groups; scalable and efficientSteiner tree based algorithms for multicast protocols.

– Yezekael Hayel – Charging the Internet without bandwidth reservation: mathematicalmodeling and analysis.

– Franck Lebeugle – Architectural integration of heterogeneous mobile technologies in-side the access network: radio link management and mobility.

2

– Helene Le Guen – Test generation: a probabilistic approach based on stochastic graphs.

– Patrick Maille – Game theory applied to Internet pricing.

– Ana Minaburo – Header compression in high speed networks. Analysis of and im-provements to the ROHC protocol.

– Francine Ngani – Combined Mobile IPv6 protocol testing with security considerations.

– Julio Orozco – QoS for multimedia flows in low-speed IP networks: active queue man-agement and marking strategies for video flows over differentiated services networks.

– Mohamed Ouarraou – Active measurements for service level specifications in the In-ternet.

– Landy Rabehasaina – Fluid queues and networks of fluid queues of the second orderin a general random environment with level dependent input and service rates.

– Franco Robledo – WAN design tools: integration of connectivity and performanceconstraints; procedures specialized for access networks and backbone networks.

– Lucian Suciu – Profile management and interface selection algorithms for mobileterminals having multiple interfaces.

– Miled Tezeghdanti – Traffic Engineering (TE) management to ensure fast rerouting incase of link failures with integration in OSPF; TE and hierarchical routing domains.

– Martin Varela – Automatic pseudo-subjective measure of the quality of multimediasignals transported over packet networks and applications.

– Rachid Zagrouba – Beyond 3G: Mobility and QoS control in all IP cellular accessnetworks.

Current position of former project-team members (including PhD studentsduring the 1999-2002 period):

• Sylvain Gombault, Assistant Professor at the ENST Bretagne (no longer in the projectbecause of changes in our focus and in the research activities at ENST Bretagne).

• PhD that were finished during the evaluation period:

– Yasser Abdelbaset: Assistant Professor, Egypt

– Nelly Barbot: Assistant Professor, ENSSAT, Lannion, France

– Atika Bousseta: Consulting Engineer, EUROGROUP, Paris

– Moulaye Hamza: in search of a position

– Jose Incera: Assistant Professor, ITAM, Mexico

– Youcef Khouaja, Engineer, France Telecom

– Octavio Medina: Temporary Engineer, ENST Bretagne

– Samir Mohamed: Temporary Engineer, IRISA

– Nathalie Omnes: Engineer, Mitsubishi, Rennes

– Olivier Paul: Assistant Professor, INT, Evry, France

– David Ros: Assistant Professor, ENST Bretagne

Last INRIA enlistments

• Bruno Tuffin was hired at the creation of the project.

3

2 Scientific aspects

2.1 Keywords:

Resource allocation, congestion control, throughput control, traffic control, traffic engineering,service differentiation, dimensioning, availability, reliability, queues, IP, interconnection, interop-erability, discrete event models, fluid flow models, multicast, multimedia, performability, metrol-ogy, performance, Markov chains, stochastic processes, protocols, QoS (quality of service), highspeed networks, network reliability, end-to-end protocols, security, simulation, dependability,pricing, Monte Carlo techniques, testing, header compression.

2.2 Research fields

The main objectives of the project are the identification, the conception and the selection of themost appropriate network architectures of a communication service, as well as the developmentof computing and mathematical tools for the fulfillment of these tasks. These objectives leadto two types of complementary research fields: the systems’ qualitative aspects (e.g. protocols’test and design) and the quantitative aspects which are essential to the correct dimensioning ofthese architectures and the associated services (performance, dependability, QoS, vulnerabilityand performability evaluation).

2.3 Project-team presentation overview

The ARMOR project works on problems related to the design and the analysis of communicationservices. Such services require functionality specifications, decisions about where and how theymust be deployed in a system, and the dimensioning of the different components of the system.The interests of the project concern not only particular classes of systems but also methodologicalaspects.

Concerning the communication systems themselves, we focus mainly on IP networks and ourconcerns go from architectural aspects to protocols, studying different aspects of the structureof networks and services: from the topological organization of nodes and links to the softwaretechniques allowing the two current versions of the IP protocol (IPv4 and IPv6) to coexist, fromthe problems related to the development of architectures allowing to provide specific Quality ofService (QoS) levels, to security or mobility aspects of the IP protocol.

Interoperability testing is essential to establish that network components interact correct-ly before they get deployed in a real environment. As such, it is considered as a part of thestandardization process. The Armor project contributes in providing solutions (methods, algo-rithms and tools) which help in obtaining efficient interoperability test suites for new generationnetworks, mainly IPv6 related protocols.

From the application point of view, our global field is IP technology in general. We areparticularly interested in the “low speed links” world, where QoS aspects are very importantand lead to many different and exciting problems (on architectural aspects, on routing, onthe protocols themselves). We also have activities in pricing methodologies (a critical area fortelecommunications providers, with many defying open problems for the near future), in manyareas related to the IPv6 technology, in the integration of packet transmission techniques intothe next generations of mobile networks, etc.

Related to the previous remarks are the quantitative aspects of most of those problems.We develop techniques for the evaluation of different aspects of the considered systems throughmodels and through measurement techniques. The quantitative aspects we are interested in areperformance, dependability, performability, QoS, vulnerability, etc.. The methods we work withgo from discrete event simulation and Monte Carlo procedures to analytical techniques, and

4

include numerical algorithms as well. Our main mathematical tools are stochastic processes ingeneral and queueing models and Markov chains in particular, optimization techniques, graphtheory, combinatorics, etc. Also in the quantitative evaluation area, we develop a methodologyable to quantify the quality of multimedia flows automatically and as humans do.

2.4 Scientific foundations

2.4.1 Introduction

The scientific foundations of our work are those of network design and network analysis. Morespecifically, this concerns the principles of packet switching and in particular of IP networks(protocol design, protocol testing, routing, scheduling techniques), and the mathematical andalgorithmic aspects of the problems, on which our methods and tools are based.

These foundations are described in the following paragraphs. We begin by a subsectiondedicated to Quality of Service, since this concept can be seen as a unifying concept of ouractivities. Then we briefly describe the specific subarea of models’ evaluation and about theparticular multidisciplinary domain of pricing problems.

2.4.2 Quality of Service

Since it is nowadays difficult to think of communication solutions dedicated to each possibleapplication, the scientific and technological communities aim towards providing general servicesallowing to give to each application or user a set of properties nowadays called “Quality ofService”, a terminology lacking a precise definition. This QoS concept takes different formsaccording to the type of communication service and the aspects which matter for a given appli-cation: for performance it comes through specific metrics (delays, jitter, throughput, . . . ), fordependability it also comes through appropriate metrics: reliability, availability; vulnerabilityfor instance in the case of WAN topologies, etc. Moreover, some aspects of QoS have subjectivecomponents: the quality of a video stream or an audio signal, as perceived by the user, is relatedto some of the previous mentioned parameters (packet loss, delays, . . . ) but in an extremelycomplex way, and with a strong subjective component.

QoS is at the heart of our research activities: we look for methods to obtain specific “levels” ofQoS and for techniques to evaluate the associated metrics. Our ultimate goal is to provide tools(mathematical tools and/or algorithms, under appropriate software “containers” or not, allowingusers and/or applications to attain some level of QoS, with an optimal use of the resources ofthe considered communication system. Obtaining a good QoS level is a very general objective.It leads to many different areas, depending on the considered systems, applications and specificgoals. Our team works on several of these areas. We can mention the wide family of routingproblems, which in Armor go from graph algorithms to routing techniques specialized to operatein the last mile part of the network under extreme performance constraints, our protocol-orientedactivities (header compression techniques, interaction between protocols, for instance betweenIPv4 and IPv6) or the research works around differentiated services. We are also concerned withspecific software engineering techniques, namely with middleware technologies in order to hideas much as possible the problems related to resource sharing, scalability and heterogeneity (forinstance, such software systems have been successfully used for stationary distributed systemsbuilt over fixed networks but they do not suit mobile settings).

2.4.3 Modeling

The scientific foundations of our modeling activities are composed of stochastic processes theoryand, in particular, Markov processes, queueing theory, graph theory, etc., either for analyticalmodels or for discrete event simulation or Monte Carlo (and Quasi-Monte Carlo) techniques.

5

We are always interested in models’ evaluation techniques for dependability and performabilityanalysis, both in static (network reliability) and dynamic contexts (depending on the fact thattime plays an explicit role in the analysis or not). We look at models from the classical so-calledcall level, leading to standard models (for instance, queueing models) and also at the burstlevel, leading to fluid models. For this more recent research field, we work both on analyticaltechniques and on discrete event simulation.

Lastly, our work on the design of the topologies of WANs leads us to optimization techniques,in particular in the case of very large optimization problems, usually formulated in terms ofgraphs. The associated methods we are interested in are composed of simulated annealing,genetic algorithms, TABU search, etc. For the time being, we have obtained our best resultswith GRASP techniques.

2.4.4 Pricing

Pricing is a good example of a multi-disciplinary research activity half-way between appliedmathematics, economy and networking. Indeed, the Internet is facing a tremendous increaseof its traffic volume. As a consequence, real users complain that large data transfers take toolong, without any possibility to improve this by themselves (by paying more, for instance).A possible solution to cope with congestion is to increase the link capacities; however, manyauthors consider that this is not a viable solution as the network must respond to increasingdemand (and experience has shown that demand of bandwidth has always been ahead of supply),especially now that the Internet has become a commercial network. Furthermore, incentives fora fair utilization between customers are not included in the current Internet.

For these reasons, it has been suggested that the current flat-rate fees, where customers paya subscription and obtain an unlimited usage, be replaced by usage-based fees. Besides, thefuture Internet will carry heterogeneous flows such as video, voice, email, web, file transfer andremote login among others. Each of these applications requires a different level of quality ofservice (QoS): for example, video needs very small delays and packet losses, voice requires smalldelays but can afford some packet losses, email can afford delay (within a given bound) whilefile transfer needs a good average throughput and remote login requires small round-trip times.Some pricing incentives should exist so that each user does not always choose the best QoSfor her application and so that the final result is a fair utilization of the bandwidth. On theother hand, we need to be aware of the trade-off between engineering efficiency and economicefficiency; for example, traffic measurements help in improving the management of the networkbut is a costly option.

2.4.5 Testing

Interoperability testing is the act of determining if end-to-end functionality between (at least)two communicating systems is as required by the base standard(s) for those systems. Confor-mance testing is the act of determining to what extent a single component conforms to theindividual requirements of the standard it is based on. In our team, we consider that confor-mance tests are used in order to validate single networks for interoperability purposes. As aconsequence, since a couple of years ago, our research activity focuses on interoperability testing.No real formal framework exists in the interoperability testing area, contrary to conformancetesting. Our purpose is to provide a formal framework (methods, algorithms and tools) forinteroperability testing which helps in obtaining efficient interoperability test suites for newgeneration networks, mainly IPv6 related protocols.

The generation of interoperability test suites is based on specifications (standards or RFCs) ofnetwork components and protocols to be tested. The model we use is an automata-like structure

6

called IOLTS (Input Output Labelled Transition Systems). It is an LTS which distinguishesinputs, outputs and internal actions.

2.5 Application domains

Our main application domains are those related to network design, both at the transport in-frastructure level and at the service level. Our expertise currently focuses on IP technology ina variety of contexts (IP QoS, IP security, IP mobility, IP telephony,. . . ), and on analysis anddimensioning tools: telecommunications architecture configuration, bottleneck search, resourceallocation policies comparison, etc. Our works on protocols and control mechanisms are alsoapplicable to other technologies besides IP, such as ATM.

Problems arising from the coexistence and interoperability of different technologies are alsoinvestigated: between IP and ATM, IP and WDM, IPv4 andIPv6, etc. In the field of traf-fic engineering and system dimensioning, technological evolution also poses a number of newperformance evaluation problems. Besides these main application domains, other importantsubjects where quantitative analysis plays a central role are, for example, the analysis of controlmechanisms, or the problems posed by pricing, which are of evident interest for operators. Inthe IP world, extensions such as mobile IP, cellular IP, security–related aspects, multicasting,and compression techniques (e.g. header compression) are also important application domains.

The first field in which the team’s expertise is in demand is that of IP networks. The usualcontext is that of an industry member who wishes to develop new techniques, or that of a userwho has to set up a new communications system or to upgrade (or more generally, modify)an existing one. This may involve a specific aspect of the system (e.g. the costs model whichallows the development of a billing policy), or a particular kind of network (for instance, ahome-network), or a family of services (for instance, a security policy).

We can also classify ARMOR’s main application domains per type of services involved.Then, the past and current expertise of the team’s members mainly involve the transport ofmultimedia flows over IP, the various network QoS management aspects, the testing techniques(interoperability tests, implementation validation tests – especially for IPv6, and test genera-tion). In this context we find, for instance, problems related to the conception of mechanismswell adapted to specific flow types and QoS goals, both at the network access level, and at theintermediary node level.

With regard to analysis and dimensioning, we contribute to the different related method-ologies (measurements, simulation, analytical techniques), and also to the development of newmathematical and software tools. We develop models for the collection of specific characteristicsof the studied systems (e.g., those related to QoS). We also develop new simulation method-ologies, in order to overcome certain limitations of the existing techniques. Finally, it shouldbe noted that networks now offer services with a certain level of redundancy, which leads toproblems of reliability. Our team has a long experience in the specific study of this systems’aspect and in related problems such as performability and vulnerability (a notion aiming atquantify the robustness of a grid without taking into account the reliability of each component).

2.6 Main contributions:

Steiner problem on networks. We have developped an original family of heuristics whichconnect the sub-trees of a Steiner tree with k-limited minimal spanning trees. The idea can beapplied on the Kruskal heuristic or on the Takahashi-Matsuyama algorithm [1]. The algorithmsneed a solution to remove the maximal not used part of the partial spanning trees [2]. Thiskind of heuristics allows the fast partial reorganization of multicast groups [3] and providesefficient routing algorithms for constrained construction of Steiner trees in all-optical networks,for example [4].

7

Interoperability testing. We have proposed a formal definition of the notion of interoper-ability, which has been considered by the testing community as the first real contribution in thisarea [5, 6]. On a pragmatic side, we validate our solutions on next generation networks. Wehave generated conformance and interoperability tests for significant IPv6 and 3GPP relatedprotocols like MIPv6 (Mobile IPv6), ROHC (Robust Header Compression), IPv4-IPv6 transi-tion mechanisms (NAT-PT, ISATAP), RIPng (Routing Internet Protocol for IPv6), etc. Thesetests have been used for many “interoperability events” such as the ETSI/Plugtest events (since2000), and the Japanese TAHI events (since 2001).

Probabilistic coverage in testing. As test suites cannot be exhaustive, test coverage con-sists in evaluating the part of the communicating system covered by a test suite. We haveproved that an extended version of Markov chains (labelled chains) can be used to master thetest campaign. We also provided a definition of reliability based on the elements really coveredduring the testing activity. See [7].

We are also involved in the “IPv6 Ready Logo Programme” which is a world wide certificationprogramme launched by the IPv6 Forum.

DSTM. The Dual Stack Transition Mechanism is a proposition submitted to the IETF in orderto solve some transition problems for remaining IPv4 hosts and applications, when the networkinfrastructure is mainly IPv6. The DSTM proposal has been widely debated and importantwork on our mechanism continues in different R&D organisations (ETRI, in Korea, implementedDSTM for Windows; HP is using it also, etc.). See www.ipv6.rennes.enst-bretagne.fr/dstm.

Diffserv. We have actively worked on DiffServ issues since the beginning of the activity of thecorresponding IETF working group. Our work has focused on the transport of multimedia flowsover the AF service class, along the following lines: proposal of a new active queue managementalgorithm [8], and packet-marking strategies for the transport of hierarchically-coded video overlow-speed, DiffServ-enabled IP networks (see our software ADserv described in page 12).

Pricing for differentiated services and congestion control. Pricing techniques for thesepurposes and from a mathematical modelling point of view, have been investigated in [9]. Amongother results, we have shown that, in order to increase his revenue, an ISP should use a priorityscheduling at his routers rather than generalized processor sharing. Optimal prices have alsobeen found for other models, such as the Cumulus [10].

Auctionning for bandwidth. This is an alternative pricing mechanism. As main contribu-tions in this area, we have improved a progressive second price scheme [11] that incentizes usersto truthfully submit their bids and that converges to an optimal allocation. The performanceof this scheme in a random environment (i.e., users entering and leaving the network) has alsobeen investigated in [12].

Monte Carlo simulation. Monte Carlo is a tool aiming at solving problems without or withvery few specific assumptions and/or large state spaces. When rare events are involved, thestandard simulation is replaced by rare event techniques such as importance sampling or impor-tance splitting. The techniques have been analyzed in [13]. They have also been implementedin SPNP, a software package developped at Duke University [14, 15, 16].

8

Quasi-Monte Carlo simulation. Quasi-Monte Carlo is a deterministic analogous of Montecarlo simulation where pseudo-random numbers are replaced by low discrepancy sequences whichfaster fill the state space. We have designed a QMC method [17] for estimating transientmeasures of Markov chains (where a direct application did not work). Error estimation, usingrandomization is (still)also an active research topic [18].

Dual processes in queueing analysis. We have discovered that the concept of dual processas proposed in Anderson’s book on Markov chains, can be useful to derive closed-forms of basicMarkovian queues, once the problem is set in discrete-time terms trhough Jensen’s method. Incase of using lattica path counting tehcniques, the dual process simplifies the combinatorial partof the analysis. See [19] and [20] for first results.

Modelling and analysis of threshold-based queues. Threshold queues with hysteresismechanisms appear naturally in the dynamic control of computer systems and telecommuni-cation networks. We have obtained a set of results on these types of models in [21] and [22](analytical work) and Stochastic Petri Net-based solutions, especially when the trafic is of dif-ferent types like in service differentiation architectures, in [23] and [24].

QQA: Quantitative Quality Assesment. We have developped a new approach able toevaluate quantitatively and automatically, and in real time if necessary, the quality of an audio,video or more generally multimedia flow. Our claim is that our approach gives to the flow avalue close to the value obtained from “average” human observers. It is based on the use ofopen queueing networks with positive and negative customers as Neural Networks, proposed byE. Gelenbe. Our method allowed us to analyze the simultaneous impact of both encoding andnetworks parameters on the quality. Representative publications for these results are [25] forvideo, [26] for audio.

Stationarity detection. In [27], we develop a very simple method to compute transientdependability measures on Markov chains. This method allows us to stop the computationwhen the stationary behavior is reached avoiding thus a considerable amount of computation.As a byproduct, we obtain the corresponding stationary measures. This method is very precisebecause the error can be specified in advance.

A new performability technique. In [28], we have obtained explicit formulae for the dis-tribution of the accumulated reward over (0, t) in a Markov chain. We also developed a veryaccurate algorithm to compute it, with a precision given in advance. The joint distribution ofall the rewards accumulated state by state has been obtained in [29].

Fluid queues. We have obtained simple formulae and precise algorithms for the distributionof the buffer content of a finite or infinite capacity fluid queue fed by a Markovian queue [30]or by an M/M/1 queue [31]. A closed form solution for tandem fluid queues fed by on-offexponential sources, with the condition that only one source is necessary to fill the first buffer,has been obtained in [32].

Stability. In [33] we have obtained the stability condition for a second order fluid queue in ageneral ergodic environment, with level dependent input and services rates, the local variancemapping being also level dependent. This was an open problem formulated as a conjecture.The stability of networks of such queues has also been obtained and has been submitted forpublication.

9

Linear service rates. We have obtained the Laplace transform of the joint distribution ofthe buffer content of all queues in a network of fluid queues with Markov modulated input ratesand linear service rates. A paper about this work is in preparation. This central result, whichis a part of a thesis, is the basis for the analysis of more general networks of fluid queues. Thetransient analysis of such a single queue has been published in [34].

Analytical modeling of TCP. We have proposed in [35] a very accurate markovian modelof TCP refining the previous works on performance evaluation of one bulk transfer TCP flowamong exogeneous traffic. While most of these works are mainly focused on the mean throughputevaluation, our model allows with low cost, a study of many other performance measures, [36],and thus a more detailed analysis of the AIMD principle, [37].

Multicast protocols. In this context, we have proposed a new protocol which uses an effi-cient method to contruct multicast trees and deliver multicast packets. Our approach is originalbecause it adopts a multicast address allocation based on source-specific channels, reduces for-warding states in non-branching nodes and implements data distribution using unicast trees.Diferent parts of this work has been proposed to several international conferences and to theIETF [38, 39]. In the same context, we have proposed an extension to the Xcast protocol [40].Our proposition is an adaptative protocol which generalizes the Xcast forwarding method andmay be parameterized to fit the size of the group. Our protocol is as efficient with small groupsas the original Xcast family protocols and it can manage more efficiently larger groups.

Multicast and MPLS. In [41, 42], MPLS LSPs are used between multicast tree branchingnodes in order to reduce forwarding states and enhance scalability. As a consequence, onlyrouters which act as branching nodes for a group need to keep forwarding states for that group.All other (non-branching) routers on the data path simply forward multicast data packets overtraffic engineered unicast routes using LSPs.

Standardization and certification, and the IPv6 Logo Programme The Armor teamhas an important activity dedicated to standardization and certification in the telecommunica-tion area. We participate to several working groups of main telecommunication standardizationinstitutes: IETF (Internet Engineering Task Force), ETSI (European Telecommunication Stan-dardization Institute), 3GPP (3rd Generation Partnership Project), etc. We are also active inthe main mailing-lists treating of new generation networks and protocols. Several proposals andcontributions to the definition of some standards and RFCs (Request For Comments) have beenpublished. They concern the IPv6-related protocols, IPv6 mobility (MIPv6), IPv4-IPv6 transi-tion mechanisms such as DSTM (Dual Stack Transition Mechanism), security (IPsec, IKE, etc.),“Universal Mobile Telecommunications System” (UMTS). Let us mention our effort on mobilityand security standards: we make an important contribution to correct protocols weaknesses1 orto improve the interaction between mobility and IPsec. We have also studied the interactionwith AAA2.

A significant result of this activity is our strong contribution to launch a world-wide cer-tification process for IPv6 products. It is the so called “IPv6 ready Logo Programme” (seehttp://www.irisa.fr/IPv6Logo/) which aims to provide means need to test existing IPv6 prod-ucts to be deployed in the market. The ARMOR team leads the technical part of this Programmeby defining the certification process itself, specifying required tests, and developping some of the

1RFC 3519 “Mobile IP Traversal of NAT Devices”2draft-dupont-mipv6-aaa-xx.txt

10

interoperability tests needed. This work is done together with the IPv6 Forum, the ETSI, theWIDE-project in Japan and the TTA (Telecommunications Technology Association) in Korea.

2.7 Project-team positioning

The ARMOR project must first be positionned with respect to both the set of INRIA projects innetworking and the set of projects dedicated to modelling. A first remark is that, by construction,ARMOR includes people working on both sides. From this macroscopic point of view, we haveat INRIA Sophie-Antipolis the tandem MISTRAL–PLANETE, with a good level of interactiontaking advantage of the same geographic situation and covering together those two aspects.

From the networking point of view, our focus on IPv6, our activity on testing, our work insecurity, and the interactions between them are specificities of the project’s activity, as topics likesatellite-based networks or interactive applications (among other) are specificities of PLANETE.We also have a strong interest in multicast routing and traffic engineering; we are putting nowforces in the home networking areas. This has points in common with the HIPERCOM project(we collaborate with them in the european project FABRIC on these topics -see 2.13) but theyfocus on wireless networking, while we look at different aspects (reservation techniques and theirperformance evaluation, perception quality and control, etc.).

From a modelling point of view, the closest INRIA project-teams are MISTRAL and HIPER-COM and, to a lesser extent TREC, RAP and PREVAL who deal with more theoretical aspectsof telecommunications networks, or who follow different approaches in the analysis of models(e.g. random geometry techniques sued by TREC). In our project, we often follow a Markovianapproach and we have strong interests in developping also numerical schemes for the effectiveevaluation of the models. Another specificity of ARMOR, which is unique at INRIA, is towork on dependability aspects and their generalizations or extensions (performability, vulnera-bility,...). We also work on Monte Carlo techniques and on Quasi-Monte Carlo. A simulationactivity already exists at INRIA (OMEGA Project), but their approach focuses on problemsinvolving stochastic differential equations, with application mainly in finance, whereas we aremore concerned with the simulation of rare events with dependability and performance evalu-ation applications. Our combination of Monte Carlo and Quasi-Monte Carlo methods is alsospecific in France.

We have also points in common with these teams, and we exploite some of them in differentcollaborating contexts: we work with MISTRAL on pricing (they participate to the ARC Prixnetheaded by Rennes), we work with the same team and many other teams inside and outside INRIAin the ARC TCP (headed by MISTRAL), as mentioned earlier we work with HIPERCOM in aneuropean project. We also collaborate with ARES (Rhone-Alpes) about IPv6 auto-configurationproblems, with TEMICS at Rennes in multimedia transport over low speed links, with R2D2(Rennes, a project working on architecture problems) in two industrial collaborations (ASIAand HADES, see 2.13).

Ouside INRIA, we are close to several other national research groups on specific aspects ofour respective activities: some examples are the PRISM group at Versailles of J. M. Fourneau,on Markovian techniques (they use different techniques in their approaches, like bounding ones),the IMAG-ID group of B. Blateau (again, they focus on other aspects like parallelism, or areinterested in specific tools, for instance, in perfect sampling approaches). Having said this, westart a collaboration with the two mentionned groups on security and dependability through anACI collaboration. We share with C. Lecot of the univ. of Savoie the interest in the specificfamily of Monte Carlo and Quasi-Monte Carlo methods. Several French teams take part in theresearch of the protocols and the algorithms for multicast communication (MASCOTTE, LRI,LIP6, LaBRI. LIRMM, ARES, CITI,...). The exchanges of information (in the G6 forums, inthe TAROT group, at the Algotel meetings) prove the complementarity of the profiles.

11

2.8 Publications

99 00 01 02PhD Thesis 0 3 5 2H.D.R (*) 0 0 0 1Journal 6 3 10 7Conference proceedings (**) 14 19 17 27Book chapter 1 0 0 2Book (written) 1 1 1Book (edited) 1Technical report 7 6 12 5Deliverable 8 7 10 6(*) HDR Habilitation a diriger des Recherches

(**) Conference with a program committee

Five main journals in which scientific staff members publish their results are: PerformanceEvaluation, Journal of Applied Probability, Telecommunication Systems, IEEE Transactions onComputers, Queueing Systems.

Five principal conferences where scientific staff members published their results on a regularbasis are ASMTA (Analytical and Stochastic Modelling Techniques and Applications), ICTS(International Conference on Telecommunications Systems), INFOCOM, IPDS (InternationalPerformance and Dependability Symposium), MCQMC (Monte Carlo and Quasi-Monte CarloMethods in Scientific Computing),

2.9 Software

2.9.1 Advanced software

ADserv. A software prototype of a DiffServ router (based on the FreeBSD operating systemand the ALT-Q software) was developed in our team. This was the first implementationof the complete DiffServ mechanism on a BSD platform, and is distributed through thefollowing web site: http://www.rennes.enst-bretagne.fr/~medina/ds-imp/.

DSTM. This transition mechanism between IPv4 and IPv6 is tested or used by differentorganisations (like the CERN). Both the documentation and the software is available fordownload at: http://www.ipv6.rennes.enst-bretagne.fr/dstm/.

ROHC. We have developed the first and, to the best of our knowledge, only existing imple-mentation of ROHC for IPv6. This implementation is not freely available, but has beenused for conformance tests and licensed to several companies.

SPNP 6.0. The Stochastic Petri Net Package is a versatile modelling tool for solving StochasticPetri Net models. It has been developed at Duke University and distributed to more than200 companies and universities. We have contributed to the integration of fluid Petri Netmodels and discrete-event simulation solutions.

2.9.2 Prototype software

BB. (Bulls and Buckets). We have developped a library allowing to build Markov models froma high level description in C or C++, based on the work done during a PhD some time ago.We intend to distribute the tool shortly through the Web.

12

DEPENDlib. We have developped a library of functions which goal is to compute depend-ability measures on models described either in a static framework or using Markov chains.The latter include procedures to evaluate accumulative metrics.

FluidSim. After two thesis in the domain, a simulator of fluid models was developed. Thesimulator allows the fast creation of the network fluid models and the analysis of thetraffic on the burst level. A server written in C++ carries out the simulation for Java GUIcustomers. We intend to distribute the tool shortly through the Web.

FluidSolve. We have developped several procedures for the computation of dependability andperformability measures and for the computation of various buffer level distributions offluid queues. A JAVA graphical interface is being developed for the modeling and analyticalanalysis of networks of fluid queues.

QMClib. We have developed in C a library implementing low-discrepancy sequences, used inQMC methods.

2.10 Collaborations

2.10.1 Collaborations with other INRIA project-teams:

• with TEMICS (Rennes): mapping multimedia flows onto a DiffServ architecture and s-tudying compression mechanisms to improve the ROHC protocol (an Associated Engineerhas been hired this month to study this topic).

• with ARES (Rhone-Alpes): behavior of auto-configuration problems (router auto-configurationprotocols); for instance, analysis of prefix stability when the network topology is changingand when nodes or network with a small mobility are introduced in the DLSAC ptotocolwe have proposed.

• with R2D2 (Rennes): see ASIA and HADES in 2.13.

• ARC Prixnet (INRIA collaborative research). Jan. 03 – Dec. 04. Responsable: ARMOR.Partners: MISTRAL project (INRIA Sophia-Antipolis), PRiSM (University of VersaillesSt-Quentin), France Telecom R&D, IBM T.J. Watson Research Center.

Field: Internet pricing. Goal: INRIA collaborative research on Internet pricing betweenactive French groups (as well as a former INRIA member now with IBM).

• ARC TCP (INRIA collaborative research). Jan. 02 – Dec. 03. Responsable: MIS-TRAL (Sophie-Antipolis). Partners: MISTRAL and PLANETE projects (INRIA Sophia-Antipolis); TREC, RAP and HIPERCOM projects (INRIA Rocquencourt); Ecole Poly-technique Federale de Lausanne (Switzerland); France Telecom R&D.

Field: analysis and performance evaluation of the TCP protocol.

2.10.2 Collaborations with French research groups outside INRIA:

• Security ACI: “Sure-Paths”. Aug. 03 – Jul. 06. Partners: ID project (INRIA Rhone-Alpes), PRiSM (University of Versailles St-Quentin).

Field: network security and reliability. Goal: to study the reliability and performabilityof systems involving large state spaces and/or rare events by means of modular decompo-sition, bounds, Monte Carlo simulation, perfect sampling.

• in VTHD and VTHD+ (see 2.13)

13

• Universite de Savoie (C. Lecot): simulation of Markov chains by QMC methods.

• We participate in the Metropolis RNRT project, headed by LIP6. This project aimsat proposing realistic models of packet behavior in IP networks and at studying metricsdefined in the corresponding IETF working group. A main objective is to develop asampling technique to restrict the number of measurements, and to determine an adequategranularity of flows allowing to obtain the relevant informations.

2.10.3 Collaboration with Foreign research groups:

WIDE project, Japan, on IPv6 and mobility.

Duke University, USA (K.S. Trivedi). Modelling and simulation of SPNs and FSPNs.

Universite de Montreal, Canada (P. L’Ecuyer). Quasi-Monte Carlo simulation methods.

IBM T.J. Watson Research Center, USA (L. Wynter). Pricing in telecommunication networksand yield management (through the Prixnet ARC).

Universidad de Los Andes, Merida, Venezuela (R. Marquez). Pricing for TCP flows (throughthe Prixnet ARC).

see BALATON project in 2.14

Our modelling approches are partly shared with international research groups whose leadersare M. Neuts (a regular visitor), G. Latouche, V. Ramaswami, K. Trivedi and we havescientific collaborations with all these people.

We cooperate with A. Krinik and his group at the CalPoly at Pomona, California, on latticepath combinatorics.

2.11 Specific hardware for experimental purpose (if relevant):

None.

2.12 Specific software for experimental purpose (if relevant):

None.

2.13 Industrial collaborations:

2.13.1 Collaborations finished in september 2003

WAN performability analysis through Monte Carlo techniques. Nov.98 – Dec. 99.ARMOR’s budget: 15 kAC. Partner: Defense Ministery.

Area: performability analysis. Goal: improvement of Monte Carlo methods previouslypublished by ARMOR, applied to large WANs in specific (military) contexts (confidentialresults).

ASIA: Accelerated Signalling for the Internet over ATM. RNRT Project. Jan.99 – Dec. 00.ARMOR’s budget: 136 kAC. Ind. partners: France Telecom, Ericsson Telecom, Airtria;academic partners: COSI Project at INRIA.

Field: QoS in the Internet. Main goal: efficient coupling between ATM and Internetprotocols and performance analysis of fair share bandwidth algorithms based on elasticABT-DT (ATM Block Transfer with Delayed Transmission).

14

RTIPA: Real Time Internet Platform Architectures (European ITEA project 99011). Oct. 99– Dec. 01. ARMOR’s budget: 312 kAC. Ind. partners: Siemens in Italy, Philips (France,Netherlands), Oratrix in the Netherlands, EolrinG, Thales Communications, France T-elecom R&D, GIP Renater in France, Ericsson Telebit in Denmark; academic partners:Politecnico di Milano (Italy), LIP6 (France), Eindhoven Univ. of Technology and CWI(Netherlands).

Field: next-generation IP networks, with support for both QoS mechanisms and IPv6.Goals: middleware development, performance evaluation of differentiated-services net-works. ARMOR in RTIPA: QoS and IPv6 issues: transport of multimedia flows, QQA(see page 9), interoperability and conformance testing; traffic engineering.

CARAT: Access control and QoS in ATM networks. Jan. 00 – Dec. 00. Partners: FranceTelecom and Celar (DGA, Defense Ministery).

Field: high speed networking. Goal: design of an access control mechanism prototype forhigh-speed networks in general, and for ATM networks in particular.

CARAHD: Access control in high speed networks. Dec. 00 – Mar. 01. This is a follow-up ofthe CARAT contract, see above.

MIRADOR Mecanisms for intrusion detection and for reacting to attacks in military domains.Oct. 99 – Dec. 01. Partners: Alcatel, CERT ONERA, Supelec and Celar (DGA, DefenseMinistery).

Field: security in networks. Goal: design of a device to detect intrusions, with militaryrestrictions (confidential results).

Reliability and vulnerability analysis with partial data. Jan. 00 – Mar. 01. ARMOR’sbudget: 10 kAC. Partner: EDF (French national electricity company).

Field: network analysis. Goal: exploration of the capability of the vulnerability theorydeveloped in the team3 to the case of partial data.

VTHD: Really Very High Speed Network. Jan. 00 – Dec. 02. ARMOR’s budget: 330 kAC. Ind.partners: France Telecom; academic partners: other teams at INRIA, at ENST Paris, atENST Bretagne, Eurocom.

Field: high speed networking. Main Goal: to create a country-wide, high-speed multi-service platform using IP over WDM technology; to experiment flow control mechanisms;to develop traffic engineering techniques in order to ensure fast rerouting in case of linkfailures. ARMOR in VTHD: platform deployment, performance measurements, IPv6 test-ing.

IPv4 IPv6 transition studies CRE project (Contrat de Recherche Externe) with FranceTelecom. Mar. 02 – Mar. 03. ARMOR’s budget: 100 kAC.

Field: IPv6. Main Goal: performance aspects of the DSTM transition mechanism (seepage 8); to study the size of the IPv4 address pool needed in production networks.

Radio coverage in 3G wireless networks. Period: 2002. ARMOR’s budget: 6.1 kAC.Partners: CRIL telecom Software, ENST Bretagne.

Field: CDMA coverage in 3G wireless Networks. Goal: CRIL Telecom Software is design-ing a software aiming at computing the coverage in UMTS networks; ARMOR’s goal wasto study and analyse the proposed algorithms, from the specifications provided by CRIL.

3A new approach to vulnerability evaluation of communication networks, S. Bulteau and G. Rubino, ITC’15,Washington, USA, Jun. 97.

15

QoS in medical information systems. Period: 2001 – 2002. Partners: TEMICS INRIAproject, Universite de Nantes, medical centers of Brest and Rennes, and the industrialmember ETIAM.

Field: QoS and information systems. Goal: to design communication tools respectingspecific QoS and security constraints in the medical area.

Cyberte: Multiple Network Interfaces Optimized Support for an IPv6 Mobile. Jan. 2002– Mar. 03; ARMOR’s budget: 174 kAC. Ind. partners: France Telecom (France), Cisco(France); academic partners: LSIIT (France).

Field: Mobile networking. Goal: Improving QoS in heterogeneous mobile network environ-ments. Managing inter- and intra-technology hand-overs in wireless networks. ARMORin Cyberte: Adaptive application development. Adapting applications’ needs to availableresources.

ASSET: Architectural Solutions for Services Enhancing Digital Television. Jun. 02 – Jul. 03.ARMOR’s budget: 141 kAC. Ind. partners: Compaq Computer (France), Thomson Broad-cast Systems (France), Dalet-A.N.N GMBH (Deutchland), INESC Porto (Portugal), Insti-tut Fuer Rundfunktechnic GMBH (Deutchland), ManageStorage International (France),SHS Multimedia (Italy).

Field: network for digital TV. Goal: middleware architecture for easy integration of dig-ital television equipment into a digital TV production system. Universal and distributedmanagement of digital TV content, efficent protocols for control of complete operationalworkflow: acquisition, editing, storage, broadcasting, archiving.

ETSI-IPv6: Support for IPv6 Interoperability. Events organized by the ETSI InteroperabilityService. Oct. 01 – Nov. 02. ARMOR’s budget: 30 kAC. Ind. partners: ETSI (EuropeanTelecommunication Standardization Institute).

Field: IPv6 interoperability testing. Main goal: IPv6 conformance and interoperabilitytest generation. Goals: organization of the IPv6 interoperability event, finding errors inimplementations and reporting to the IETF.

NGNI: IPv6 interoperability and QoS testing. Dec. 01 – May 02; ARMOR’s budget: 25 kAC.Ind. partner: 6WIND (France).

Field: IPv6 interoperability testing. Main goal: IPv6 benchmarking and promotion inEurope. Goals: launching of the IPv6 certification program now called the “IPv6-readyLogo Program”.

2.13.2 Collaborations active in september 2003

Probabilistic test algorithm. Period: 2002–2005. ARMOR budget: 90 kAC. Partner:ALITEC.

Field: testing. Goal: design and evaluation of a probabilistic method of software testing.

NGDG: New Generation Distributed Gateway. Jun. 02 – Jul. 04. ARMOR’s budget: 326 kAC.Partner: ALCATEL Stuttgart.

Field: high performance access network. Goal: optimized architecture and protocols forthe control of a versatile, dependable access multi-gigabit network: auto-configuration,topology and resource discovery process, fast rerouting, route optimization, and routeprotection.

16

FABRIC: Federated Applications Based on Real-time Interacting Components. Sept. 02 –Feb. 03. ARMOR’s budget: 32 kAC. Ind. partners: Philips (Netherlands), Thomson(France); academic partners: Eindhoven Univ. of Technology (Netherlands), TNO Physicsand Electronics Lab. (Netherlands), Maelarden Univ. (Sweden), Scuola Superiore S. Anna(Italy), Univ. College (UK) and CSEM (Switzerland).

Field: home networking. Main goal: middleware developments, identification of userneeds, performance evaluation for home networking. Second goal: preparation for a sec-ond project (STREPS class; code: STRUCTURE) about home networking for 2004–2007.ARMOR in FABRIC: modelling activities; in STRUCTURE: IPv6 issues, modelling, reser-vation techniques, QQA (see page 9).

Ubique: QoS Profile Management and Interface Selection Algorithm for Multi-Interface Mo-bile Terminals. Jun.02 – Jun. 04. ARMOR’s budget: 166 kAC. Partner: France Telecom.

Field: Mobile IP. Main goal: to design a QoS profile management and an interface selectionalgorithm.

HADES: High Throughput and Security. Sept. 01 – Sept. 03. ARMOR’s budget: 60 kAC. Ind.partners: AQL, SAGEM, SILICOMP Reseaux, SOREP; academic partners: R2D2 INRIAproject, INT Evry.

Field: security analysis in high speed networks. Main goal: study and performance e-valuation of encoding protocols and security architectures for high-speed networks (workperformed for the Defense Ministery).

IDsA: DNSsec Infrastructure and Applications. Sep. 02 – Jul. 04. ARMOR’s budget: 218 kAC.Partners: France Telecom, AFNIC.

Field: network security. Goal: DNS security, authentification and integrity of DNS re-quests, efficient key management, deployement of DNSsec to achieve dynamic networksecurity, IPv6 deployement, 0-configuration, opportunistic encryption.

2.13.3 Starting collaborations

Let us mention two starting industrial and academic collaborations, playing an important rolein part of our objectives for the next period:

• CARAMEL. CARAMEL is a newly labelized RNRT project. Its goal is to study auto-configuration mechanisms for home networking. Different aspects of auto-configurationwill be taken into account, like layer 3 configurations, to allow the construction of somecomplex topologies due to the heterogeniety of transmission media or resources. It will alsocover the organization of the interface between the ISP and the home network. Partners:NPG – Laboratoire LSR-IMAG, Thomson Multimedia R&D France, France Telecom R&D,JUNIPER NETWORKS FRANCE, Legrand S.A., 6WIND, Herve Schauer Consultants,Alcatel CIT.

• COSINUS. This newly labelized RNRT project will study the behaviour of the ROHCheader compression mechanism in a real UMTS platform, in the PDPC layer. It will alsofind the adequate parametrization of this complex protocol, in order to optimize the radioresources. In collaboration with the INRIA project TEMICS, we will study the impact ofthe residual tramission error on multimedia flows, in order to evaluate the QoS perceivedby the user. Partners: Alcatel CIT, EURECOM, France Telecom, Thales Communication.

17

2.14 Other funding, public, European, regional, ...:

ARC Prixnet. Total budget: 55 kAC.

ACI security: Sure-Paths. ARMOR’s budget: 83 kAC.

We participate to two Actions specifiques of the STIC CNRS Dep., one on performance eval-uation, the other one on mobility.

BALATON . Jan. 00 – Dec. 01. Total budget: 10 kAC. Partner: Department of Telecommu-nication (Technical University of Budapest, Hungary)

Field: Stochastic fluid models and multicast routing problems with maximal probabilityof the QoS.

We participate in the NoE Euro3Gi.

2.15 Teaching

All members of the team except three are teachers in three different institutions, the university ofRennes 1 and two engineering schools: ENST Bretagne and INSA Rennes. The Rennes INRIAUnit is closely related to the doctoral school of Rennes 1 in Mathematics and in ComputerScience. We participate in the teaching activities of the corresponding DEAs, we receive animportant part of the students during their internships and after that, to prepare their PhDs.

• University of Rennes 1. The ARMOR members that belong to this institution are F. An-dre, B. Cousin and R. Marie (Professors), L.-M. Le Ny and C. Viho (Associate Profes-sors). Their main teaching activities and responsabilities during the evaluation periodare summarized here: responsabilities of several diplomas (“Bac+5” level, that is, lastunder-graduate year) in Computer Science, different courses in all the Computer Scieceformations in Rennes 1 and in the area (DEA, DESS, Eng. Diploma of IFSIC, INSA,ENST Bretagne, ENSAI, Supelec, IUT Rennes).

• J.-M. Bonnin, F. Dupont, L. Toutain and D. Ros are Assistant Professors at the ENSTBretagne. Their main teaching activities and responsabilities during the evaluation periodwere the following: responsability of different specializations at the last year of the ENSTBretagne, responsability of a Master between Mexico (ITAM) and France; courses at ENSIand Sup’Com Tunis, Tunisia, at Cocodi University of Abidjan, at South East Universityof Nanjing (China).

• INSA Rennes. M. Molnar is Assistant Professor at the Department of Computer Science,INSA. His teaching activities during the evaluation period are all in Computer Science atINSA and also at ENSAI.

• Main teaching activities of the three INRIA researchers of the team:

– G. Rubino: Graduate level in Computer Science (network analysis) and in Probability(Markov chains); last year in Engineering schools (performance evaluation); one-weekcourses on Performance Evaluation given at Buenos Aires, Cairo, Mexico, Montevideoand, on a regular basis, at Beyrouth; 80 hours/year except last year (40 hours), andexcluding the courses abroad.

– B. Sericola: Graduate level in Computer Science (dependability analysis); last yearsin Engineering schools (performance evaluation); 70 hours/year.

– B. Tuffin: Under-graduate level in Computer Science (queuing); 70 hours/year.

18

2.16 Visibility

Edition activities in journals: Performance Evaluation, Naval Research Logistics, IEEECommunications Magazine

Edition responsability of the Engineering Techniques Series (Hermes, France)

Editoral responsability at the Pedagogical Collection of the GET (national Group of Engineer-ing schools in Telecommunications)

R. Marie and G. Rubino are members of the WG7.3 (Performance Evaluation).

International Program Committees: 3rd Int. Conf. on Numerical Solution of Markov Chains(Raleygh, USA, 99); PNPM’99 and PNPM’01 (Petri Nets and Performance Models); PDP-TA’99, PDPTA’00 and PDPTA’02 (Parallel and Distributed Processing Techniques andApplications); ESS’99 (11th European Simulation Symposium); Icil’99 (Int. Conferenceon Industrial Logistics); TOOLS’2000 (11th Int. Conference on Modelling Techniques andTools for Computer Performance Evaluation) and TOOLS’2002; IPDS’2000 (4th IEEE In-t. Computer Performance and Dependability Symposium); SIGMETRICS’2000 and 2002(ACM Int. Conf. on Measurement and Modeling of Computer Systems); IFIP ATM &IP 2000 (8th IFIP Workshop on Performance Modelling and Evaluation of ATM and IPNetworks); MMR’2000 (2nd Int. Conf. on Mathematical Models in Reliability Theory);ISCC’2000 and ISCC’2002 (5th and 6th IEEE Symposium on Computers and Commu-nications); ASMTA’2000, ASMTA’2001, ASMTA’2002, ASMTA’03 and ASMTA’04 (7ththrough 11th Int. Conf. on Analytical and Stochastic Modelling Techniques); ITC’17 (17thInt. Teletraffic Congress); V Int. Conf. on Operations Research; CARI’2002 (African Con-ference on Computer Science Research); DSN’04 (5th Int. Conf. on Dependable Systemsand Networks); MAM5 (5th Int. COnf. on Matrix Analytical Methods).

National Program Committees: DNAC’2000 (New Architectures in Telecommunications), AL-GOTEL’2000 (2emes Rencontres Francophones sur les Aspects Algorithmiques des Telecommunications),NOTERE’2000 (3eme Colloque International sur les Nouvelles Technologies de la Repartition).

Other organizing activities: organization of schools (Summer School RHDM’99 in Networking;French-Mexicain School in Networks and Distributed Systems, sept. 2003); organizationof “‘Thematic days” at INRIA Rennes (on IPv6 in 2000, on IP telephony in 2003).

European activities as experts for the evaluation of IST proposals on networking.

Strong participation at the IETF on several topics related to IPv6 (security, mobility, routing,metrology), at ETSI also on IPv6.

Membership to the IPv6 Task Force France, in connexion with G6.

3 Main evolution of the objectives during the evaluation period

3.1 Planned objectives

The ARMOR project was created at the end of 1999. The main goal was to put together peoplecoming from the modelling area with people working on networking. This objective was relatedto the association between GET, through the ENST Bretagne) and INRIA (through the RennesUnit), instanciated in our team.

The specific objectives from the scientific point of view were to focus our research work inproblems related to the control of heterogeneous contexts in IP and ATM networks, for instance,

19

problems associated with access control mechanisms, with bandwidth management, etc. Asecond specificity of our planned goals was to integrate quantitative aspects in our work, takinginto account the background of the modelling part of the team. Last, several methodologicalaspects were part of the announced objectives, in the area of the analysis of stochastic models(stochastic graphs, Markov processes, queueing systems). It must be said that the team hasa long history in dependability evaluation of complex systems, and a specific objective was tomaintain our research effort in that direction.

3.2 Main evolution

The main evolution in our objectives during the evaluation period was the concentration of ournetworking interests around the IP technology. Indeed, the number of works concerning ATMnetworks decreased rapidly (see the activity reports of the team in the period).

A second movement clearly visible in the traces of our work is the strong focus on the IPv6version of the protocol (our team is today a major actor in France on the different aspects ofthis technology4. We also enlarged our areas of interest in this domain, for instance, integratingnew activities around testing (interoperability and conformance); see page 10.

A third and recent evolution is the focus on problems associated with mobility and next-generation mobile networks. This is due to the evolution in the same direction of our partner, theENST Bretagne, and the possibility of developing an independent cooperation on these themes.The reinforcement of ARMOR’s efforts is a preliminary step in that direction.

Another recent and significant evolution is the investment on pricing, an exciting multidis-ciplinary area with many unsolved and critical problems. We have now a stable and visibleresearch effort in the area, and we also have in mind the possibility of developing another inde-pendent activity in this direction.

4 Objectives for the next four years

We have decomposed our objectives in different areas, trying to make explicit the relationsbetween them and between the main results obtained in the evaluation period (see 2.6).

4.1 Performance and Dependability Evaluation of Networks

The solution techniques developed in the dependability analysis of computing systems (in partic-ular, in our team) can be very useful for the quantitative evaluation of communication networks.That is why we keep a close eye on the recent results obtained in the dependability domain.Actually, the very important area of networks security, for instance, although conceptually farfrom the dependability notion, can be very close to the evaluation viewpoint, and the modelingmethods can be inspired one from the other.

The different points that we would like to develop are listed in the following items.

• One of the objectives on analytical methods is to obtain a better understanding on theproperties and use of dual processes for the analysis of queuing systems. They appear tobe a new way of obtaining closed-form expressions in some basic cases. The next step willbe to identify the limits of this approach.

• We should be able to extend the results obtained for stationary tandem fluid queuesin a Markovian environment to some more complex networks composed of such queues.

4For instance, F. Dupont is the author of the first public domain IPv6 stack; we host one of the nodes ofthe French experimental IPv6 network; we belong to the team proposing one of the solutions for enabling thecoexistence of the two versions of the protocol; etc.

20

Nevertheless, the resulting complexity can make us imagine solution techniques basedon bounds computation of the performance measures we are interested in. This kind ofapproach, which has been already used in the dependability domain giving very goodresults, should be applicable also to the evaluation of networks of fluid queues.

• The modeling of transmission control protocols under various scheduling services is veryimportant in order to represent faithfully the behavior of the Internet network. Thesestudies must also evolve with the notion of services differentiation to which the Internetseems to tend. It is particularly important to combine the modeling of protocols, suchas TCP, with active queue management such as EMD (Early Message Discard) or RED(Random Early Discard) allowing the network to discard packets waiting in a queue beforeits saturation. From a modeling viewpoint, we believe that fluid models can be particularlyuseful, to avoid that each packet is associated with an event and so, to obtain reasonablecomputation times for the evaluation of performance measures.

• Simulation is particularly important for the evaluation of networks. It is often the lastresort when the analytical models do not apply or are prohibitive in terms of complexity.During the past, we made several developments in the domain of systems dependabilityevaluation. We think that these works can also be useful to the evaluation of networks atthe burst level, that is using fluid queues.

• Few works have dealt with the quantitative evaluation of security in networks and we thinkthat it is possible to develop methodology in this direction, supported in particular by thestochastic Petri net mathematical tool. The task is hard because we have not only tomodel the behavior of different types of intrusions but also to formally define the relevantsecurity evaluation measures we are interested in.

• Wireless Local Area Networks (WLANs) of the IEEE 802.11 type represent at presentgreat hopes in telecommunications. Different variants of this standard exist today, andothers are under study. One of these variant proposes the implementation of the QoSin the MAC (Medium Access Control) layer. It is particularly important to model andevaluate the performance of existing QoS management methods in order to identify theirstrength and weaknesses and, thus, to propose other methods more efficient.

4.2 On QQA

Concerning our technology in the QQA activity (see page 9), our main objectives are the fol-lowing:

• the development of a real–time multimedia quality control protocol, to be integrated intoteleconferencing, streaming and VoIP applications,

• the integration of quality assessment and quality control for home networking applications,

• the development of an industrial partnership to transform our methodology into effectivesolutions for real users.

• Finally, we are interested in studying mapping strategies for H.264 video over DiffServ AFnetworks. H.264-coded sequences will be marked using different schemes and transportedover a simulated DiffServ network under several load conditions. Subjective quality assess-ment of the disturbed sequences will be used to train a random neural network in orderto obtain a quality assessment tool that mimics human perception using only networkparameters and other readily available data. Such a tool will then allow us to choose the“best” mapping strategy for a given scenario.

21

We believe that the technology we have developed can be used in other contexts, and wehave two specific objectives for the next period on these lines:

• the development of a tool allowing a user to allocate the amount of bandwidth exactlyneeded over the next short period of time (say, over the next few minutes), based on theuse of the same RNN tool but dedied to traffic analysis and prediction,

• and the development of utility functions as perceived by customers for evaluating newpricing schemes for network services, again following the same approach of our QQAactivity, and based on the same RNN tool.

4.3 On network management

Following our current research (efficient management of high speed network) we plan to enlargeour research field toward management of more specific networks (e.g. dynamic and Ad Hocnetworks, WDM links, radio links, etc). Our expertise on fast rerouting, multicast routing,constraint-route computation, route optimization, could be adapted to these different networkswith their distinctive features. For example, we plan to take into account (a) the splitting ofequipments during multicast tree building processes, (b) new constraints induced by signal lossesin radio links during routing processes, (c) the consequences of multiple interconnected networksduring topology discovery processes (d) security in dynamic networks (e) fault-tolerant routesin Ad Hoc networks.

Also, some of our present results in this area deserve to be transfered toward industry:enhancement of DNS Security and efficient control of access networks. We will develop ouractivity into the Internet standardization body to promote our solutions and protocols.

4.4 On Home Networking and IPv6

We forecast that the introduction of networks in an home environment will lead to new researchsubjects, either in the development of new solutions, like routers auto-configuration protocols, orin the adaptation of existing protocols, for instance those created to manage quality of service.This activity can be viewed as the continuation of the research on IPv6 realized in the teamduring the previous years. The standardization of this new version of the IP protocol is almostover. Most of operating systems and router manufacturers include an IPv6 stack. New domainsfor networking and applications are rising, like Home Networking or Mobile telephony. Thesedomains will require a large amount of addresses. Contrary to business networks, where theIEEE 802.x family of technology is dominant, the size of the network in home networking islimited to a house or an apartment, resulting in a complex network topology since differenttechnologies can be used to transport information (IEEE 1394 bus may be used to connectHi-Fi and video devices, power line may be used to control equipments like light bulb, shadesand Wireless network or Bluetooth may be used to connect computers, printers and PDA).Moreover, the topology of the network may evolve dynamically; for example, Bluetooth linkdevices may leave or enter the network. The home network may also be multi-homed, sinceseveral accesses (GPRS, ADSL, satellite TV) managed by different ISP may be available at thesame time. Auto-configuration has been presented as one of the attractive new feature of IPv6.If this mechanism simplifies the network management on administrated networks, it cannot becompared to a real plug and play network since routers still need to be manually configured.

We thus have started to study router auto-configuration. A protocol based on the knowledgeof the network (similar to the link state protocol used by IGP) completes auto-configurationmechanisms already developed for IPv6 for hosts. This protocol should allow partitioning ormerging of networks, limited mobility, but scalability in a first step will not be a key issue.

22

We also plan to study the behavior of DSTM, the transition mechanism we develop, to allow aauto-configuration of IPv4 equipments, if an IPv6 network exists.

Television will be one of the first applications to benefit from home networking. Currently, inthe Customer Equipment world, video is mostly transmitted directly on links, like DV directlyon an IEEE 1394 bus. This allows a management of QoS property of the link, but limitsthe scope of the video to a limited set of devices. Interconnection, with at least PersonalComputer using mostly Ethernet or WiFI network is an important issue. The introductionof the IP layer can ease the interconnection between these different medias but the quality ofservice management has to be solved. The IETF has studied for a long time the QoS aspects,but with a limited impact on backbone where overprovisionning allows more flexibility. Forhome networking, overprovisioning is not always possible: some network technology can offervery limited bandwidth, but on the other hand, the number of interconnection equipment, asthe number of flow requiring QoS, is also very limited. Network Calculus theory will be moreefficient to compute reservation parameters. Another difference comes form the market approachof networking. For communication between computers, the network has to be simple enoughand all the complexity is pushed to the edge where computers can correct the errors introducedby the network. In a mass market approach, end equipment like television have to be as cheapas possible to be competitive. This limits the amount of memory available to reduce the jitterintroduced by the network. The complexity will be supported by interconnection equipmentslike bridges and routers.

4.5 On Diffserv

The access network (ADSL, UMTS) generally acts as the bottleneck. We plan to focus ondifferent methods to manage it so that multimedia flows will nto be disturbed by data flowswhen congestion occurs. We will study the mapping of multimedia flows on DiffServ classes ofservices. This study has already been initiated the team and should be continued to adapt to newtypes of coding like MPEG-4, where granularity is higher. We will also strenghten our studies onActive Queue Management to improve the use of the link and ease the parameterization. This isalso an important issue on UMTS radio links where the bandwidth varies dynamically with thequality of the transmission. In that field we have a very good experience on Header Compressionfor IPv6/UDP/RTP flows, and have already proposed mechanisms to adapt the compression tofeedback coming from the radio layer. We will pursue this investigation to integrate multicastand unidirectional flows.

4.6 On testing

The testing activity has the following main objectives:

• Based on the results achieved in the formal definition of the notion of interoperability, wewill provide a complete formal framework (methods, algorithms and tools) for interoper-ability testing.

• We also plan to apply the previous framework to new generation networks, mainly to IPv6related protocols and Mobile Networks. A challenge linked to this work is to move thepragmatic Internet community towards a more formal approach of interoperability testing.By this way, our involvement in the standardization process (like the “IPv6 Ready LogoProgram”) will be increased.

• Test coverage consists in evaluating the part of the communicating system covered by atest suite. A way to answer to this problem is to study the functional coverage and its

23

reliability. Some interesting results on Markov chain could be used to foresee the way toobtain an efficient test suite, for example by using visiting probabilities.

4.7 On pricing

We plan to strengthen our network pricing activity in the next four years. This requires furthermathematical modeling and analysis of different possible pricing schemes and their mathematicalcomparison. Second, all pricing models include a ”utility function” representing the servicevaluation by the users. In the literature, stringent assumptions are imposed on this functionfor the sake of convenience but that might not represent the actual demand and lead to non-optimal prices. Based on tests and in collaboration with the QQA activity, we plan to estimatethis function based on ”real-world” data, that is, considering subjective measures from realusers. This would require extensive testing since we do not need the mean opinion score, butrather the whole distribution with respect to some specified parameters.

To the best of our knowledge, most of the work on network pricing reported in the literatureis of a theoretical nature. However, the deployment of pricing schemes on a real network arisessome interesting technological problems, for instance: (i) The design and implementation ofprotocols for dynamic pricing (for example, protocols for tariff distribution and negotiation). (ii)The integration of pricing and metrology tools, that may be needed to deploy complex usage-based pricing schemes. (iii) The integration of pricing methods into a service differentiationarchitecture (like the IETF’s DiffServ). Therefore, we also intend to put these methods inpractice, to test their feasibility and to compare them on an experimental platform in order toselect the most likely to be implemented at large scale as a compromise between efficiency andsimplicity.

Finally, up to now this activity has focused on wired networks, but we aim at extendingit to wireless and ad hoc networks, dealing both with theoretical modeling issues and withpractical and implementation issues like in the case of wired networks. Indeed, third generationof wireless networks, such as the UMTS network, will have to deal with demanding applicationsin terms of bandwidth and QoS requirements. Even if the capacity has increased with respectto the networks of second generation, demand is supposed to be so important that congestionpricing will probably be the most relevant means to control and manage the traffic (for instancewatching a movie on a terminal would require the capacity of an entire base station). Followingthe same lines, there must be in an ad hoc network some incentives to forward traffic of othernodes. Without associated pricing scheme, choosing to not participate in the network may be inthe (short-term) interest of a node, meaning to not expend energy without receiving any directgain from doing so. Pricing is then a concept that will introduce incentives for collaborationinto the architecture. Also, in the current situation, owners of private WiFi networks oftenchoose to encrypt their networks to prevent outsiders from accessing them. Many of these basestations could be used to allow Internet access to a much larger set of users. Pricing wouldbe a mechanism to incentivize owners of existing private wireless base stations to open theirnetworks to the public, as well as incentivize institutions to deploy base stations in order toobtain ubiquitous WiFi coverage.

References

[1] M. Molnar. Construction of more advantageous multicast diffusion trees. In InternationalConference on Dynamic Control and Systems (DYCONS’99), Ottawa, Canada, August1999.

[2] R. Marie and M. Molnar. Maximal removal forest (invite). In SAPM’00: Symposium onAdvanced Performance Modeling 2000, Orlando, novembre 2000.

24

[3] M. Molnar and R. Marie. Algorithmes de gestion dynamique d’un groupe de communicationmultipoint. In Algotel’2001, Saint Jean de Luz, mai 2001.

[4] A. Zsigri, A. Guitton, and M. Molnar. Two multicast algorithms for sparse splitting capablenetworks. In 7th Optical Network Design & Modelling Conference (ONDM’2003), Budapest,Hungary, february 2003.

[5] S. Barbin, L. Tanguy, and C. Viho. Towards a formal framework for interoperability testing.In M. Kim, B. Chin, S. Kang, and D. Lee, editors, 21st IFIP WG 6.1 International Con-ference on Formal Techniques for Networked and Distributed Systems, pages 53–68, ChejuIsland, Korea, august 2001.

[6] C Viho. On the importance of interoperability testing events. In Invited paper for theWorkshop session of TestCom’2002 Conference, Berlin, march 2002.

[7] H. Le Guen and Marie. R. Visiting probabilities in non irreducible markov chains withstrongly connected components. International Journal of Simulation: Systems, Science &Technology. Special Issue on Modelling and Simulation of Computer Systems, 3(3-4).

[8] J. Orozco and D. Ros. An Adaptive RIO (A-RIO) queue management algorithm. InProceedings of QoFIS 2003, number 2811 in Lecture Notes in Computer Science. Springer,2003.

[9] B. Tuffin. Charging the internet without bandwidth reservation: an overview and bibliog-raphy of mathematical approaches. Journal of Information Science and Engineering, 19(5),2003. To appear.

[10] Y. Hayel and B. Tuffin. A mathematical analysis of the cumulus pricing scheme. TechnicalReport 1466, IRISA, July 2002.

[11] B. Tuffin. Revisited progressive second price auctions for charging telecomunication net-works. Telecommunication Systems, 20(3-4), July 2002.

[12] P. Maille and B. Tuffin. The progressive second price mechanism in a stochastic environ-ment. Netnomics, 5(2):119–147, 2003.

[13] H. Cancela, G. Rubino, and B. Tuffin. Mttf estimation by monte carlo methods usingmarkov models. Monte Carlo Methods and Applications, 8(4):312–341, 2002.

[14] K.S. Trivedi and B. Tuffin. Implementation of importance splitting techniques in stochasticpetri net package. In B.R. Haverkort, H.C. Bohnenkamp, and C.U. Smith, editors, Com-puter Performance Evaluation: Modelling Tools and Techniques, volume 1786 of LectureNotes in Computer Science, pages 216–229. Springer Verlag, 2000.

[15] C. Hirel, K.S. Trivedi, and B. Tuffin. Spnp version 6.0. In B.R. Haverkort, H.C. Bohnenkam-p, and C.U. Smith, editors, Computer Performance Evaluation: Modelling Tools and Tech-niques, volume 1786 of Lecture Notes in Computer Science, pages 354–357. Springer Verlag,2000.

[16] K.S. Trivedi and B. Tuffin. Importance sampling for the simulation of stochastic petri netsand fluid stochastic petri nets. In HPC’2001: High Performance Computing, pages 228–235,Seattle, USA, avril 2001.

25

[17] C. Lecot and B. Tuffin. Quasi-monte carlo methods for estimating transient measures ofdiscrete time markov chains. In Proceedings of the fifth International Conference on MonteCarlo and Quasi-Monte Carlo Methods in Scientific Computing, Lecture Notes in Statistics,Singapoure, November 2002. Springer Verlag.

[18] L.-M. Le Ny and B. Tuffin. Parallelisation d’une combinaison des methodes de montecarlo et quasi-monte carlo et application aux reseaux de files d’attente. RAIRO: RechercheOperationnelle, 34(1):85–98, 2000.

[19] A. Krinik, G. Rubino, D. Marcus, R. J. Swift, H. Kasfy, and H. Lam. Dual processes tosolve single server systems. Journal of Statistical Planning and Inference (a paraıtre), 2003.

[20] M. L. Green, A. Krinik, C. Mortensen, G. Rubino, and R. J. Swift. Transient probabilityfunctions. a sample path approach. DMTCS (a paraıtre), 2003.

[21] L.-M. Le Ny. Exact analysis of a threshold-based queue with hysteresis and a delayed ad-ditional server. In 9th International Conference on Telecommunications Systems: Modelingand Analysis, Dallas, USA, mars 2001.

[22] L.-M. Le Ny and B. Tuffin. A simple analysis of heterogeneous multi-server thresholdqueues with hysteresis. In Proceedings of Applied Telecommunication Symposium (ATS),San Diego, USA, April 2002.

[23] L.-M. Le Ny and B. Tuffin. Modeling and analysis of threshold queues with hysteresisusing stochastic petri nets: the monoclass case. In Proceedings of IEEE Petri Nets andPerformance Models, pages 175–184, Aachen, Germany, 2001. IEEE CS Press.

[24] L.-M. Le Ny and B. Tuffin. Modeling and analysis of multi-class threshold-based queueswith hysteresis using stochastic petri nets. In Petri Nets 2002, Lecture Notes in ComputerScience. Springer Verlag, 2002.

[25] S. Mahevas and G. Rubino. Bound computation of dependability and performability mea-sures. IEEE Transactions on Computers, 50(5):399–413, 2001.

[26] S. Mohamed, G. Rubino, and M. Varela. Performance evaluation of real-time speech througha packet network: a random neural networks based approach. Performance Evaluation (aparaıtre), 2003.

[27] B. Sericola. Availability analysis of repairable computer systems and stationarity detection.IEEE Transactions on Computers, 48(11), novembre 1999.

[28] B. Sericola. Occupation times in markov processes. Communications in Statistics - Stochas-tic Models, 16(5), 2000.

[29] M. Bladt, B. Meini, M. Neuts, and B. Sericola. Distribution of reward functions oncontinuous-time markov chains. In 4th International Conference on Matrix Analytic Meth-ods in Stochastic Models (MAM4), Adelaide, Australia, Juillet 2002.

[30] B. Sericola. A finite buffer fluid queue driven by a markovian queue. Queueing Systems,38(2), 2001.

[31] N. Barbot and B. Sericola. Stationary solution to the fluid queue fed by an m/m/1 queue.Journal of Applied Probability, 39(2), juin 2002.

[32]

26

[33] L. Rabehasaina and B. Sericola. Stability analysis of second order fluid flow models in astationary ergodic environment. Annals of Applied Probability (a paraıtre).

[34]

[35] S. Fortin and B. Sericola. A model of tcp in wide area networks. In 10th IEEE/ACMInternational Symposium on Modeling, Analysis and Simulation of Computer and Telecom-munication Systems (MASCOTS 2002), Fort Worth, USA, octobre 2002.

[36]

[37]

[38] A. Boudani and B. Cousin. Sem: A new small group multicast routing protocol. In10th International Conference on Telecommunications (ICT’2003), Tahiti, Papeete, France,fevrier 2003.

[39] A. Boudani and B. Cousin. Simple explicit multicast (sem). draft IETF: draft-boudani-simple-xcast-01.txt, juin 2001.

[40] A. Boudani, A. Guitton, and B. Cousin. Gxcast: Une generalisation du protocole xcast. InPremier Congres Francophone, Majecstic’03, Marseille, Octobre 2003.

[41] A. Boudani and B. Cousin. A new approach to construct multicast trees in mpls networks.In Seventh IEEE Symposium on Computers and Communications (ISCC02), Taormina,Italy, juillet 2002.

[42] A. Boudani and B. Cousin. The mpls multicast tree (mmt). draft IETF: draft-boudani-mpls-multicast-tree-02.txt, novembre 2001.

27