Marc Barthelemy

08 avril 2010

Most transportation networks are embedded in space and consequently, their topology is not independent from spatial constraints. In particular, these constraints can induce a hierarchical organization with hubs controlling specific regions of space, non-trivial correlations between the weights, the connectivity pattern and the actual spatial distances of vertices, and the emergence of anomalous fluctuations in the betweenness-degree correlation function. In this talk I will illustrate these effects both from an empirical and modeling point of view, with various examples ranging from the large scale air travel network to smaller scales of inter- and intra- cities transportation networks.

Mathieu Bastian

25 mars 2010

In this practical talk, I will present 0.7 version of Gephi, open-source graph visualization & manipulation software. Gephi possesses a complete set of network visualization and analysis features, proposes major innovations in dynamic and hierarchical graphs and above all is a flexible and extensible architecture for plugins. I will firstly sum up embedded features: Layout, Metrics, Ranking, Filters, Partition, Preview, Clustering, DataLab. Then, I will present and discuss dynamic and hierarchical graphs features and roadmaps while showing live demo. Afterwards, I will outline our approach by presenting the Gephi project, some insights about performance and the various possibilities of using and extending Gephi in the way scientists and engineers can rely on. To conclude I will speak about plugin development and show how to create a plugin in five minutes.

Jean-Jacques Pansiot

11 mars 2010

Slides

Active and passive measurements for topology discovery have known an impressive growth during the last decade. If a lot of work has been done regarding inter-domain topology discovery and modeling, only a few papers raise the question of how to extract intra-domain topologies from measurements results. In this paper, based on a large dataset collected with mrinfo, a multicast tool that silently discovers all interfaces of a router, we provide a mechanism for retrieving intra-domain topologies. The main challenge is to assign an AS number to a border router whose IP addresses are not mapped to the same AS. Our algorithm is based on probabilistic and empirical IP allocation rules. The goal of our pool of rules is to converge to a consistent router to AS mapping. We show that our router-to-AS algorithm results in a mapping in more than 99% of the cases. Furthermore, with mrinfo, point-to-point links between routers can be distinguished from multiple links attached to a switch, providing an accurate view of the collected topologies. Finally, we provide a set of large intra-domain topologies in various formats.

Pierre Borgnat

04 mars 2010

Slides

Dans un but de classification du trafic émis par des ordinateurs via internet mais aussi de détection des anomalies de trafic, une méthode de caractérisation des communications d’ordinateurs est étudiée. Un espace de représentation des motifs de connexions de chaque ordinateur est étudié en ce qu’il représente le trafic échangé, la dispersion de ce trafic, la connectivité de l’ordinateur. On montrera que cette représentation permet par exemple de quantifier plus finement les graphlets empiriques utilisés par la méthode BLINC (Karagiannis et al. 2005). Cette représentation du trafic échangé permet alors de mettre en place une classification non supervisée du trafic internet par une approches de classification par MST (Minimum Spanning Tree), sur des liens backbones et sans disposer du contenu des paquets ni du trafic bidirectionnel (ici validé sur le trafic backbone d’un lien transpacifique opéré par WIDE, Japon).

Marc Lelarge

25 février 2010

Slides

Nous introduisons un modèle de diffusion qui généralise à la fois le processus de contact et la percolation ‘bootstrap’. Nous analysons ce processus sur des graphes aléatoires dilués. Ceci nous permet de retrouver des résultats connus (taille de la composante géante, seuil de percolation) et nouveaux (condition de cascade, impact de différentes vaccinations). Les preuves reposent sur des idées de couplages développées récemment par Janson et Luczak.

Jean-Baptiste Rouquier

28 janvier 2010

Slides

In many networks, vertices have hidden attributes that are correlated with the networks topology. For instance, in social networks, people are more likely to be friends if they are demographically similar. In food webs, predators typically eat prey of lower body mass. We explore a setting in which the networks topology is known, but these attributes are not. If each vertex can be queried, learning the value of its hidden attributes but only at some cost then we need an algorithm which chooses which vertex to query next, in order to learn as much as possible about the attributes of the remaining vertices. We assume that the network is generated by a probabilistic model, but we make no assumptions about the assortativity or disassortativity of the network. We then query the vertex with the largest mutual information between its type and that of the others (a well-known approach in active learning) or with the largest average agreement between two independent samples of the Gibbs distribution which agree on its type. We test these approaches on two networks with known attributes, the Karate Club network and a food web of species in the Weddell Sea. In several cases, we found that the average agreement algorithm performs better than mutual information, and both perform better than simpler heuristics. The algorithms appear to explore the network intelligently, first querying vertices at the centers of communities, and then vertices along the boundaries between communities.

Damien Magoni

02 février 2011 de 14h à 15h : salle 26-00/228

Slides

In order to untie entities, applications and devices, CLOAK introduces a P2P overlay architecture and redefines the notion of a session. A session is a communication descriptor that contains everything needed for linking entities, applications and devices together in a flexible way. A session represents the identity and the management information of a given communication. Thus the lifetime of a communication between several entities is equal to the lifetime of its corresponding session. A device can move or be changed for another without terminating the session. Similarly, an application can be changed for another if deemed appropriate or even moved (i.e. mobile code) also without terminating the session. Finally entities can move or change (i.e. be transferred to another entity) without terminating the session if this is appropriate for a given communication. We can see that in our new architecture, entities, applications and devices are loosely bound together during a communication and all the movements and changes of devices, applications and entities are supported. To be able to implement our notion of session, we need to introduce several new layers of identifiers in order to decouple lower layers (devices) from middle layers (entities) and from higher layers (applications). Hence the name CLOAK given to this project. We add naming layers above the existing ones in order to give abstract names to objects such as entities and devices. Thus we lay a cloak on the current Internet architecture.

Marcelo Dias De Amorim

14/01/2010, de 11H00 à 12H00, en salle 847

Slides

In this talk, we will discuss some preliminary work on the spatial characterization of intermittently-connected wireless networks. The motivation is that the literature has focused almost exclusively on the temporal aspect of contacts and inter-contacts between nodes. Furthermore, it is frequently assumed that contacts have infinite capacity, which is a too strong consideration especially in wireless networks. We propose the surround indicator as a metric to exhibit the spatial dimension of contacts in opportunistic networks. This metric can be used for example as a measure of potential interference that a contact might undergo or to understand the validity of forwarding strategies like betweenness centrality. We evaluate the surround indicator on two existing datasets. Our preliminary results reveal that contacts have too heterogeneous surrounds to be considered only in terms of duration. This work is part of Nadjet Belblidia’s thesis and is done in collaboration with Jérémie Leguay (Thalès), Vania Conan (Thalès), Jon Crowcroft (Cambridge University), and Serge Fdida (UPMC).

Christophe Prieur

19/11/2009, de 11H00 à 12H00, en salle 555

Slides

En dix ans, la vision Google des réseaux sociaux a cédé la place à la vision Facebook. Dans le milieu des années 90, avec l’apparition du web et son intégration progressive dans la société, s’est répandue l’image d’un monde connecté, village global traversé par les autoroutes de l’information permettant de relier n’importe qui à n’importe qui d’autre en six degrés de séparation. En 1998, Google apportait à cette civilisation-là un algorithme permettant de traiter le gigantesque bazar que constituait déjà le graphe des pages web, permettant d’en extraire les pages les plus pertinentes pour le genre humain. Aujourd’hui, les résultats de Google sont un acquis, c’est la wikipédie de tout un chacun, presque aussi poussiéreuse que le Larousse de la grand-mère. Je sais, quand j’en ai besoin, chercher des informations dans le réseau des réseaux, reste donc à trouver des informations quand je ne les cherche pas (Google ne me dira ce qu’est Paf le chien que si j’ai l’idée de le lui demander), et pour cela, écouter ce qui se passe dans… mon réseau. Parce que si je suis à O(6) degrés de séparation de n’importe qui, je préfère réduire la portée de mon réseau pour limiter la cacophonie, et le pagerank ne m’aidera pas pour ça. Nous allons donc profiter de notre expérience de vieux googliens pour aborder les réseaux personnels avec des outils que les anciens analystes des réseaux sociaux n’avaient pas, quitte à ce qu’ils nous reprochent de leur piquer leur jouet.

Fabien de Montgolfier

12/11/2009, de 11H00 à 12H00, salle 847

Slides

La vidéo à la demande (VoD) est un service en pleine expansion. Avec le développement des modems ( box ) d’accès à l’ADSL, offrant maintenant des capacités de stockage, on peut imaginer un service totalement décentralisé. Les films sont pré-chargés sur certains modems. Les utilisateurs se connectent aux modems qui les ont pré-chargés et visionnent, sans intervention d’un serveur (d’où un service robuste et à coût faible). Nous montrons que le nombre de films disponibles peut croitre de façon linéaire (ou quasi-linéaire sous d’autres hypothèses) en fonction du nombre de modems et que toutes requêtes d’utilisateurs peuvent être satisfaites AFP. Travail réalisé avec Y. Boufkhad, F. Mathieu, D. Perino et L. Viennot.