EN FR
EN FR


Section: Partnerships and Cooperations

National Initiatives

ANR BottleNet

Participants : Isabelle Chrisment [contact] , Thibault Cholez, Vassili Rivron, Paul Andrey, Quentin Rouy.

The Quality of Experience (QoE) when accessing the Internet, on which more and more human activities depend, is a key factor for today’s society. The complexity of Internet services and of user’s local connectivity has grown dramatically in the last years with the proliferation of proxies and caches at the core and access technologies at the edge (home wireless and 3G/4G access), making it difficult to diagnose the root cause of performance bottlenecks. The objective of BottleNet is to deliver methods, algorithms, and software systems to measure end-to-end Internet QoE and to diagnose the cause of experienced issues. The result can then be used by users, network and service operators or regulators to improve the QoE.

The ANR BottleNet project (https://project.inria.fr/bottlenet) started this year with a kick-off on the 1st of February 2016. It involves many partners in the field of computer networks and QoE: Inria Muse and Diana teams, Lille1 University, Telecom Sud-Paris, Orange, IP-Label. The objective of BottleNet is to deliver methods, algorithms, and software systems to measure Internet QoE and diagnose the root cause of poor Internet QoE. Our goal calls for tools that run directly at users’ devices. We plan to collect network and application performance metrics directly at users’ devices and correlate it with user perception to model Internet QoE, and to correlate measurements across users and devices to diagnose poor Internet QoE. This data-driven approach is essential to address the challenging problem of modeling user perception and of diagnosing sources of bottlenecks in complex Internet services. BottleNet will lead to new solutions to assist users, network and service operators as well as regulators in understanding Internet QoE and the sources of performance bottleneck.

Our first research question was to evaluate the impact of web advertisement on users' QoE. An interdisciplinary approach was developed at MADYNES, by which we extend the common notion of “quality of information“ on free news websites (usually based on journalistic content) to a notion of quality of experience for the user, that takes into account the degraded delivery of information by the multiplication of third party contents. We implemented a measurement tool as a web browser extension and made a dataset by browsing many news websites accessed with and without ad-blockers. The first statistical results retrieved from the dataset show that web-advertisement has a huge negative impact on QoE, for example multiplying the mean page load time by more than one order of magnitude and increasing the variance even more, while adblockers' profiles show faster and more uniform performances. These results have to be further refined but already show that web-advertisement, and more generally third-party content provider, play a huge role in poor Internet QoE and that it is a key parameter to investigate in the project. This study is leading to a structural analysis of the ad regulation mechanisms in the field of web journalism. Adblockers not only upgrades the QoE of visitors, but also contributes to define what “acceptable ads“ should be.

This year the following task have been completed:

  • Development of a platform to collect QoS & QoE on french news websites (Quentin Rouy, Telecom Nancy student). The first exploratory data collecting campaign measured the impact of Web Advertisement on client QoS, using FatHom plugin for Firefox (a tool from MUSE/Inria, partner of the BottleNet ANR project).

  • Implementation of statistical treatment schemes (Paul Andrey, ENSAE student) to correlate QoS, economic models and adblocking effects on news websites.

  • Preparation of a systematized measurement campaign on french and international news sites, in order to publish to a large audience.

ANR Doctor

Participants : Thibault Cholez [contact] , Thomas Silverston [contact] , Xavier Marchal, Cédric Enclos, Elian Aubry, Daishi Kondo, Olivier Festor.

The DOCTOR project http://www.doctor-project.org is an applied research project funded by the French National Research Agency (ANR), grant <ANR-14-CE28-000>, and supported by the French Systematic cluster. The project started on December 2014 for three years. It involves five partners specialized in network monitoring and security: Orange Labs (lead), Thales, Montimage, Université de technologie de Troyes and LORIA/CNRS. The DOCTOR project advocates the use of virtualized network equipment (Network Functions Virtualization), to enable the co-existence of new Information-Centric Networking stacks (e.g.: Named-Data Networking) with IP, and the progressive migration of traffic from one stack to the other while guaranteeing the good security and manageability of the network. Therefore in DOCTOR, the main goals of the project are: (1) the efficient deployment of NDN as a virtualized networking environment; (2) the monitoring and security of this virtualized NDN stack.

We presented the whole project at the IRTF Information-Centric Networking Research Group (ICNRG) in January.

This year, we made contributions in three critical points for the deployment of virtualized NDN network: security, performances and interoperability. First, we identified a critical vulnerability in the NDN protocol design that allows an attacker to perform efficient DoS attacks [46] by either self-answering to his own requests or answering to clients before the server. We proposed several remediation strategies to this problem.

On the performance topic, we designed and implemented a tool similar to Iperf, Ndnperf [22](http://madynes.loria.fr/software/ndnperf_cpp.zip), that can measure the maximum throughput of a program serving NDN Data. We identified critical limitations that can harm real-time services (live streaming, VOIP, etc.), and proposed several recommendations and improvements that can increase the throughput up to 8 times when combined together.

Finally, we also designed and implemented an HTTP/NDN gateway that can be used to transport web content on an NDN network, thus benefiting from its caching and multicast properties while being totally transparent for the client and the server [47]. Those three contributions were published and demonstrated in the main conference of the domain: ACM ICN.

PIA LAR

Participants : Kévin Roussel, Ye-Qiong Song [contact] .

LAR (Living Assistant Robot) is a PIA (Projet investissement d'avenir) national project getting together Inria (MAIA and MADYNES projects), Crédit Agricole (lead), Diatelic and Robotsoft. The aim is to develop an ambient assisted living system for elderly including both sensors and assistant robots. The task of Madynes team is the development of a WSN-based system integrating both sensors of the environment and sensors and actuators embedded on a mobile robot. The research issues include the QoS, energy and mobility management.

This project has ended in March 2016. Some new results are obtained including the use of Cooja simulator for RIOT OS based WSN simulation and an in-depth analysis of some timing inaccuracy problems introduced by MSPSim which is an emulator of MSP430 MCU [27]. A synthesis of our achievements on LAR project is reported in the PhD thesis of Kévin Roussel (http://www.theses.fr/196570603).

FUI HUMA

Participants : Jonathan Arnault, Giulia de Santis, Pierre-Olivier Brissaud, Jérôme François [contact] , Abdelkader Lahmadi, Isabelle Chrisment.

The HUMA project (L’HUmain au cœur de l’analyse de données MAssives pour la sécurité) is funded under the national FUI Framework (Fonds Unique Interministerial) jointly by the BPI (Banque Publique d'Investissement) and the Région Lorraine. It has been approved by two competitive clusters: Systematic and Imaginove. The consortium is composed of three academic (ICube, Citi, Inria) and five industrial (Airbus Defence and Space, Intrinsec, Oberthur, Wallix, Sydo) partners. The leader is Intrinsec.

This project targets the analysis of Advanced Persistent Threat. APT are long and complex attacks which thus cannot be captured with standard techniques focused on short time windows and few data sources. Indeed, APTs may be several months long and involve multiple steps with different types of attacks and approaches. The project will address such an issue by leveraging data analytics and visualization techniques to guide human experts, which are the only one able to analyze APT today, rather than targeting a fully automated approach.

In 2016, our contribution focused on defining a clustering technique in order to group individual events into a common one. We applied our technique to darknet data as shown in section 6.2.1. In addition, we also start the modeling of an attacker process by considering the first phase of APT, i.e. the reconnaissance phase by analyzing scanning activities using Hidden Markov Model in section 6.2.1. We also technically contribute to the definition of APT scenarios by providing a very stealthy scanning approach (Wiscan described in 6.1.2). Finally, from a project management point of view, Inria is in charge of leading the work-package related to data analytics technique for analyzing security probe events.

Inria-Orange Joint Lab

Participants : Jérôme François [contact] , Rémi Badonnel, Olivier Festor, Maxime Compastié, Paul Chaignon.

The challenges addressed by the Inria-Orange joint lab relate to the virtualization of communication networks, the convergence between cloud computing and communication networks, and the underlying software-defined infrastructures. This lab aims at specifying and developing a GlobalOS (Global Operating System) approach as a platform or a software infrastructure for all the network and computing resources required by the Orange network operator. Our work, started in November 2015, concerns in particular monitoring methods for software-defined infrastructures, and management strategies for supporting software-defined security in multi-tenant cloud environments.

CNRS-INS2I PEPS NEFAE

Participants : Thibault Cholez [contact] , Wazen Shbair, Isabelle Chrisment, Jérôme François.

The need to monitor the increasing proportion of HTTPS traffic while preserving the privacy of users led us to propose a privacy-preserving monitoring framework that allows efficient identification of encrypted traffic (based on full TLS sessions), without relying on any decryption (no HTTPS proxy). It is based on a new set of well-tuned network features to characterise the service inside the encrypted traffic and on machine learning algorithms. The CNRS PEPS founded NEFAE project aims to specifically address the practical challenges toward real time identification of encrypted traffic by developing a next-generation firewall prototype.

This year we first built and made publicly available a new HTTPS dataset(http://betternet.lhs.loria.fr/datasets/https/) (with complete raw data) so that researchers can compare their identification algorithms. We also improved our HTTPS monitoring framework to allow real-time identification of HTTPS services with only a few data packets instead of the full TLS session. We show better performances that the related work in all dimensions: better accuracy, earlier decision and more fine-grained identification). A running prototype is also under development to evaluate the scalability and overhead of our solution.

CNRS-INS2I PEPS SURF

Participants : Abdelkader Lahmadi [contact] , Jérôme François, Isabelle Chrisment.

The SURF project, funded by the CNRS PEPS program, addresses the challenge of a developing a methodology for the joint modelling and the analysis of the Cyber security and the safety of industrial systems in the context of the factory of the future. The project involves partners from the Heudiasyc Laboratory of the University of Technology of Compiègne (UTC), the CRAN laboratory and the Inria Madynes team. The goal of the project is to make a joint effort from safety and cyber security communities to address the challenges of a joint modelling of industrial systems while including attacks, vulnerabilities and failures. During the year 2016, with the partners of the project, we have mainly identified the key challenges regarding this issue where we identified the common models, metrics and analysis methods that should be built. We have also organized a scientific day (http://surf.loria.fr) with many industrials (EDF, PSA and Sentryo) and academic to share with them our work and clearly identify the requirement and experience regarding this issue. This short term project is ended by this year, however a consortium is established for further long term projects (ANR, FUI or H2020) to address the identified challenge of a joint analysis of the cyber security and the safety of industrial control systems.

ANR FLIRT

Participants : Olivier Festor [contact] , Rémi Badonnel, Thibault Cholez, Jérôme François, Abdelkader Lahmadi, Laurent Andrey.

FLIRT (Formations Libres et Innovantes Réseaux & Télécom) is an applied research project leaded by the Institut Mines-Télécom, for a duration of 4 years. It includes 14 academic partners (engineering schools including Telecom Nancy), 3 industrial partners (Airbus, Nokia Group and Orange), 2 innovative startups (the MOOC agency, and Isograd), as well as 3 professional or scientific societies (Syntec Numérique, Unetel, SEE). The project objective is to build a collection of 10 MOOCs (Massive Open Online Courses) in the area of networks and telecommunications, 3 training programmes based on this collection, as well as several innovations related to pedagogical efficiency (such as virtualization of practical labs, management of student cohorts, and adaptative assessment). The Madynes team is leading a working group dedicated to the building of a MOOC on network and service management. This MOOC will cover the fundamental concepts, architectures and protocols of the domain, as well as their evolution in the context of future Internet, and will include practical labs and exercises using widely-used tools and technologies.

Technological Development Action (ADT)

ADT UASS

The goal of this ADT is while still providing assistance in developing the Aetournos platform to help in the UAV Challenge Medical Express. Through this ADT, funded by Inria, Raphaël Cherfan has coordinated students work on the platform and tutoring the Aetournos team for the 2016 Outback Joe Search and Rescue / Medical Express Challenge, and help in the design and buidling of a novel Hybrid UAV.

ADT VERTEX

This ADT started on 2016 and will end on 2018. The Madynes project is a major partner funded at the level of 120k€. ADT VERTEX buildt upon the foundations of the Grid'5000 testbed aims to reinforce and extend it towards new use cases and scientific challenges. Several directions are being explored: networks and Software Defined Networking, Big Data, HPC, and production computation needs. Already developed prototypes are also being consolidated, and the necessary improvements to user management and tracking are also being performed.

ADT COSETTE

This ADT started on 2013 and is endind on 2016. The Madynes project is the only partner funded at the level of 120k€. ADT COSETTE, for COherent SET of Tools for Experimentation aims at developing or improving a tool suite for experimentation at large scale on testbeds such as Grid'5000. Specifically, we will work on (1) the development of Ruby-CUTE, a library gathering features useful when performing such experiments; (2) the porting of Kadeploy, Distem and XPFlow on top of Ruby-CUTE; (3) the release of XPFlow, developed in the context of Tomasz Buchert's PhD; (4) the improvement of the Distem emulator to address new scientific challenges in Cloud and HPC. E. Jeanvoine (SED) is delegated in the Madynes team for the duration of this project. A subsequent project is planned to start at the end of 2016 (ADT SDT).

ADT RIOT

RIOT ADT is a multi-site project with Infine and Madynes teams, which started in December 2016 for a duration of two years. The high-level objective is to (1) contribute open source code, upstream, to the RIOT code base, (2) coordinate RIOT development within Inria, with other engineers and researchers using/developing RIOT, (3) coordinate RIOT development outside Inria, help maintain the RIOT community at large (see http://www.riot-os.org and http://www.github.com/RIOT-OS/RIOT) which aims to become the equivalent of Linux for IoT devices that cannot run Linux because of resource constraints.

This year MADYNES team has mainly contributed to the efficient MAC layer protocol implementation issues. We have built a general MAC protocol module (gnrc mac module) for providing critical development tools for MAC protocol developers in the RIOT community (https://github.com/RIOT-OS/RIOT/pull/5941; https://github.com/RIOT-OS/RIOT/pull/5942; https://github.com/RIOT-OS/RIOT/pull/5949; https://github.com/RIOT-OS/RIOT/pull/5950; https://github.com/RIOT-OS/RIOT/pull/6069; https://github.com/RIOT-OS/RIOT/pull/6072). Based on these generic functions, we first contributed to the functionality and performance improvement of an universal example MAC protocol (Lw-MAC) (https://github.com/RIOT-OS/RIOT/pull/5941). We then implemented iQueue-MAC, which is a robust, energy efficient and traffic adaptive MAC protocol (https://github.com/RIOT-OS/RIOT/pull/5618). Currently, we have finished to implement most of the designed features of iQueue-MAC, such as the low duty-cycle scheme, the adaptive slots allocation scheme and the multi-channel operation. Experimental results collected from samr21-Xplained-pro boards showed that iQueue-MAC is robust and has a extremely low packet drop ratio, even when interference is strong.

Inria Project Lab

IPL BetterNet

Participants : Isabelle Chrisment [contact] , Thibault Cholez, Vassili Rivron.

The Inria Project Lab BetterNet (https://project.inria.fr/betternet) launched in October 2016. Its goal is to build and deliver a scientific and technical collaborative observatory to measure and improve the Internet service access as perceived by users. We will propose new original user-centered measurement methods, which will associate social sciences to better understand Internet usage and the quality of services and networks. Tools, models and algorithms will be provided to collect data that will be shared and analyzed to offer a valuable service to scientists, stakeholders and civil society.

The Madynes team leads the IPL and in particular Isabelle Chrisment who coordinates the project. Several actions have already been done over the first months:

  • Organization of the Kick-Off the 19th November in Paris;

  • Recruitment of a shared PhD with SPYRALS (Inria/University of Lille3) in order to develop probes and collecting platform;

  • Servers installation in LHS (High Security Laboratory) for the hosting of the different BottleNet and BetterNet data collection and opendata platforms;

  • Preparation of small and middle scale QoE and QoS data collection with users. Conception of incentives and rewards (value added services).