Vidéos en ligne

Les Grandes Conférences du LIG - The LIG Keynote Speeches
 - 
LIG
 -  07 février 2019

Trading off memory for bandwidth in a content-centric Internet

Jim ROBERTS

Some 96% of Internet traffic is currently generated by the transfer of digitized content, directly from content providers (CPs) like Netflix and Google, or via content delivery networks (CDNs) like Akamai and Limelite. In such a content-centric network there is clearly scope for significant economies in the cost of infrastructure by trading off memory for bandwidth. By retrieving popular content items from local storage, requirements for upstream bandwidth are greatly reduced. These economies are imperfectly realized in the current network where there is a mismatch between the objectives of network operators on one hand, and major CPs and CDNs on the other. The latter tend to jealously protect their profitable business models, notably by encrypting content delivery and thus preventing operators from transparently caching popular items at advantageous sites. They also have little incentive to cooperate in optimizing infrastructure costs through proactive placements, as long as their customers experience adequate quality. The talk will discuss how the Internet is likely to change to more effectively deal with its content-centric demand. Our analysis is based on mathematical models developed to determine cache hit rates accounting for observed characteristics of content popularity. These models enable a quantification of the memory for bandwidth tradeoff and an evaluation of alternative network structures. Our conclusion is that the future Internet will deliver the vast majority of content from datacenter-equipped central offices at the edge of the core network, or from caches located even closer to users in the access network. Most content delivery will still be controlled by major CPs and CDNs and we discuss how the network infrastructure owner will be able to persuade them to realize optimal placements through an appropriate pricing scheme.

Lire la suite
Évènement
 - 
GENCI / GRICAD
 -  17 janvier 2019

Hackathon GENCI 2018 à Grenoble

GRICAD

Le hackathon HPC est un évènement coordonné par GENCI et qui s’est tenu en décembre 2018 dans différents mésocentres de calcul intensif français. Son objectif était de réunir des équipes d’étudiants, d’ingénieurs et de chercheurs autour de sujets issus de réelles problématiques de recherche portant sur des codes logiciels de calcul intensif (portage, optimisation, implémentation de nouvelles fonctionnalités, etc.). A Grenoble, deux équipes, encadrées par GriCAD, ont participé à cet exercice sur deux sujets : l’un portant sur un code utilisé en physique des particules , l’autre portant sur un code proposant une méthode aux éléments discrets pour la micro-mécanique (logiciel nommé YADE). Il est à noter que cette dernière équipe a été désignée lauréate de l’exercice dans la catégorie « optimisation ».

Lire la suite
Les Grandes Conférences du LIG - The LIG Keynote Speeches
 - 
LIG
 -  10 janvier 2019

Privacy-preserving aggregation of data from multiple sources

David POINTCHEVAL

Gigabits of data are regularly aggregated in order to deliver statistics and recommendations, or even to make decisions. These data are processed in clear by many providers that offer valuable services, but at the cost of a huge risk with respect to privacy. The providers themselves or even hackers could exploit these data for malicious purposes. Privacy-by-design would be preferable.
Cryptography has recently developed new tools in order to allow aggregation on encrypted data, with fully homomorphic encryption and functional encryption. However, whereas they work well for one user, they fail to aggregate data that come from different sources, in particular when these sources do not trust each other.
In this talk, we will present new techniques of aggregation for data that come from multiple mutually distrustful sources, so that privacy is guaranteed, and the data owners keep control on the performed aggregation.

Lire la suite
Journée thématique
 - 
EcoInfo
 -  26 novembre 2018

L'obsolescence, dans tous les sens : Introduction

Françoise BERTHOUD

Retouvez l'intégralité des vidéos la conférence EcoInfo : L'obsolescence dans tous les sens

Lire la suite
Les Grandes Conférences du LIG - The LIG Keynote Speeches
 - 
LIG
 -  08 novembre 2018

40 years of static analysis of numerical programs

Nicolas HALBWACHS

Static analysis of programs consists in extracting guaranteed properties about all executions of a program without executing it. Such properties are useful in compilation, verification, optimization and evaluation of programs. Abstract interpretation, introduced by Patrick and Radhia Cousot in the late seventies, is the theoretical framework of static analysis. In this talk, we will focus on static analysis of numerical properties, like variable boundedness or more general invariant relations between numerical variables. During the last decades, such analyses have been widely studied, in view of finding a compromise between the expressiveness of considered properties and the cost of the analysis. We will try to summarise these works together with their main applications.

Lire la suite
Les Grandes Conférences du LIG - The LIG Keynote Speeches
 - 
LIG
 -  04 octobre 2018

Majority judgment: Why it should be used to rank and elect

Rida LARAKI

Every well-known voting system in use today hides important vices that can deny the will of the electorate including majority vote with only two candidates (the domination paradox), approval voting, all methods that ask voters to compare candidates (i.e., rank-order them), and point-summing methods. The underlying reason: the inability of voters to adequately and honestly express their opinions. Majority judgment asks voters to evaluate every candidate in an easily understood common language of ordinal grades such as: Great, Good, Average, Poor, or Terrible. Majorities determine the electorate’s evaluation of each candidate and the ranking between every pair of candidates (necessarily transitive), with the first-placed among them the winner.

Lire la suite
L'Atelier Sciences et Voix
 - 
GIPSA-LAB
 -  21 juin 2018

La voix des enseignants : usages et prévention

Maëva GARNIER
Les Grandes Conférences du LIG - The LIG Keynote Speeches
 - 
LIG
 -  07 juin 2018

The Cognitive Packet Network - Reinforcement based Network Routing with Random Neural Networks

Erol GELENBE

The Cognitive Packet Network (CPN) is an experimental network routing protocol which uses specific Quality of Service (QoS) objectives incorporated in a Goal Function, together with network measurement by Smart Packets (SPs). It updates neural network based Oracles in routers using Reinforcement Learning, in order to dynamically select network paths so that end users can convey their payload traffic with a performance that matches the Goal as closely as possible. The Goal can include conventional QoS metrics such as delay and loss, as well as Real-Time objectives, as well as newer metrics of interest including Energy Consumption and Security. Payload traffic is forwarded using source or segment routing, selected through the reinforcement learning approach, while SPs conduct their exploration using a node by node process by seeking the best direction from each Oracle. CPN has been implemented in various contexts: on 10-40 node test-beds, on an intercontinental scale as an overlay network, within SDN routers, and as a means to convey task requests over the Internet to Cloud servers. Our presentation will detail the CPN algorithm and the Random Neural Networks that are used to implement the Oracles. We will also present relate experimental measurements and results. The work has appeared in a variety of journals and conferences including CACM, Proceedings IEEE, IEEE J. Sel. Areas in Comms.

Lire la suite

Pages