|
|
Soutenance de thèse
The defense will be held in Room S907
Orange Labs, France Telecom,
38 rue du General Leclerc
92794 ISSY-LES-MOULINEAUX
The committee will be composed by:
Marco AJMONE-MARSAN, Professor at Politecnico di Torino
Serge FDIDA, Professor at Université Pierre et Marie Curie
Holger KARL, Professor at Universität Paderborn
Daniel KOFMAN, Professor at Telecom ParisTech
Luca MUSCARIELLO, Research Scientist at Orange Labs
James ROBERTS, Research Scientist at INRIA
Dario ROSSI, Professor at Telecom ParisTech
Title : “Traffic and Resource Management in Content-Centric Networks: Design and Evaluation”
The advent of the World Wide Web has radically changed Internet usage from host-to-host to service access and data retrieval. The majority of services used by Internet’s clients (The web, Video on Demand, IP Television, on-line gaming, social networking) are content-centric in the sense that they focus on named content objects rather than on host-location. Contrarily to its usage, the original Internet revolves around host-to-host communication for which it was conceived. Even if Internet has been able to address the challenges offered by new applications, there is an evident mismatch between the architecture and its current usage.
In this context, many projects in national research agencies propose to re-design the Internet architecture around named data. Such research efforts are commonly identified under the name of Information Centric Networking (ICN). This thesis focuses on the Content-Centric Networking proposition. We first analyze the CCN communication model with particular focus on the bandwidth and storage sharing performance. We compute closed formulas for data delivery time, that we use in the second part of the thesis as guideline for network protocol design. Second, we propose some CCN congestion control and forwarding mechanisms. In particular, we present a first window based receiver driven flow control protocol named Interest Control Protocol (ICP). Furthermore, we introduce a hop-by-hop congestion control mechanism in order to obtain early congestion detection and reaction. We also extend the original ICP congestion control protocol implementing a Remote Adaptive Active Queue Management mechanism (RAAQM) in order to efficiently exploit heterogeneous (joint/disjoint) network paths. Finally, we introduce a distributed forwarding mechanism that bases its decisions on per prefix and per interface quality measurement without impacting the system scalability. To support our results and test the proposed mechanisms, we have also published the source code of a CCN Packet Level simulator (CCNPL-Sim) under the GNU GPLv2 license.
Soutenance de thèse
Lieu : Amphithéâtre Emeraude à Télécom ParisTech (46, rue Barrault – 75013 Paris)
Membres du jury :
Rapporteurs :
Directeur de thèse
- Laurent PAUTET, Professeur Télécom ParisTech
- Marc GATTI, Directeur Technique Thales Avionics
Examinateurs :
- Elie NAJM, Professeur Télécom ParisTech
- Bernard LECUSSAN, Professeur ONERA
- François PECHEUX, Professeur LIP6 Paris 6
- Jean-Loup TERRAILLON, Directeur Technique ESA
Title : “Modélisation de Plate-Forme Avionique pour Exploration de Performance en Avance de Phase”
Facing a growing complexity, embedded systems design relies on model-based approaches to assist the refinement of system requirements, to contribute to a better allocation of software and hardware resources, and to proceed to early analysis. Following these principles, model-based estimation of system performances is of great interest in order to anticipate on future integration issues. A key aspect when exploring platform performances is the usage of the hardware resources.
Current model-based approaches approximate (i) hardware components characteristics by set of predefined properties corresponding to a general category of component, and (ii) interactions between components in terms of distribution over time.
We present a flexible modeling and simulation framework, to describe with different levels of detail dimensioning characteristics of simulation: precision of the description of the execution platform and distribution of interactions. Our method relies on two standardized languages: AADL to quickly model an avionic system and perform different types of analysis (e.g. schedulability), and SystemC to refine the description of the execution platform (OS and hardware), simulate this latter and evaluate its performances. We expose the mapping rules we defined to generate a SystemC model from the execution platform model described in AADL, and demonstrate the effectiveness of our process by deploying it on avionics use-case.
Soutenance de thèse
Lieu : Laboratoire LINCS, Salle du Conseil (4éme étage) au 23 Avenue d’Italie, 75013, Paris
Membres du jury :
Rapporteurs :
Directeur de thèse
- Dario ROSSI, Maître de Conférences Télécom ParisTech
Examinateurs :
- Isabelle DEMEURE (Télécom ParisTech)
- Tijani CHAHED (Télécom SudParis)
- Fabien MATHIEU (INRIA)
Invité :
- Timur FRIEDMAN (Université Pierre et Marie Curie)
Title : “On the congestion control of Peer-to-Peer applications: the LEDBAT case”
In the last years, Internet delays are considerably growing, causing a performance deterioration of interactive applications. This phenomenon is getting worse with the increasing popularity of bandwidth-intensive applications, as video streaming, remote backup and distributed content dissemination systems. The cause of these delays has been identified with the excess buffering inside the network, called bufferbloat. Research efforts in this direction head toward active queue management techniques and end-to-end congestion control. However, an ultimate solution to thisproblem is yet to come.
In this context, we investigated LEDBAT, a low-priority delay-based transport protocol introduced by BitTorrent. This protocol is specifically designed to transfer large amount of data without affecting the delay experienced by other applications or users.
First, we analysed transport-level performance of LEDBAT by means of experimental measurement, simulation and analytical model. Specifically, we evaluated LEDBAT as it is, comparing its performance to standard TCP or to other lower-than best effort protocols. We then identified a later-comer advantage affecting the original proposal. We tackled this unfairness issue proposing fLEDBAT, which re-introduces intra-protocol fairness maintaining the original LEDBAT objectives.
As a second step, we studied the impact of the LEDBAT protocol on BitTorrent performance. More in details, through simulations and real network experiments, we analysed how BitTorrent impacts on the buffer occupancy of the access node. BitTorrent performance was evaluated in terms of completion time, which can be considered as the cardinal metric to assess the user quality of experience. In both scenarios, results showed that LEDBAT decreases the completion time with respect to standard TCP. This is an unexpected phenomenon, to which we find a preliminary, though not conclusive, explanation. Moreover, LEDBAT significantly reduces the buffer occupancy, that translates in lower delays experienced by competing interactive applications.
Soutenance de thèse
Lieu : Amphithéâtre JADE à Télécom ParisTech (46, rue Barrault – 75013 Paris)
Membres du jury :
Rapporteurs :
Directeur de thèse
- Pierre Senellart, Maître de Conférences Télécom ParisTech
Examinateurs :
- Elie NAJM, Télécom ParisTech
- Julien MASANÈS, Internet Memory Foundation
- Marc TOMMASI, Université Lille 3
Title : “Deriving Semantic Objects from the Structured Web”
This thesis focuses on the extraction and analysis of Web data objects, investigated from different points of view: temporal, structural, semantic. We first survey different strategies and best practices for deriving temporal aspects of Web pages, together with a more in-depth study on Web feeds for this particular purpose.
Next, in the context of dynamically-generated Web pages by content management systems, we present two keyword-based techniques that perform article extraction from such pages. Keywords, either automatically acquired through a Tf−Idf analysis, or extracted from Web feeds, guide the process of object identification, either at the level of a single Web page (SIGFEED algorithm), or across different pages sharing the same template (FOREST algorithm).
We finally present, in the context of the deep Web, a generic framework which aims at discovering the semantic model of a Web object (here, data record) by, first, using FOREST for the extraction of objects, and second, by representing the implicit rdf:type similarities between the object attributes and the entity of the Web interface as relationships that, together with the instances extracted from the objects, form a labeled graph. This graph is further aligned to a generic ontology like YAGO for the discovery of the graph’s unknown types and relations.
|
|
