Dans les prochaines années, l’Internet des objets (IoT) constituera l’une des applications générant le plus de données. Actuellement, les données de l’IoT sont stockées dans le Cloud. Avec l’augmentation du nombre d’objets connectés, la transmission de la grande quantité de données produite vers le Cloud génèrera des goulets d’étranglement. Par conséquent, les latences seront élevées. Afin de réduire ces latences, le Fog computing a été proposé comme un paradigme étendant les services du Cloud jusqu’aux périphéries du réseau. Il consiste à utiliser tout équipement localisé dans le réseau (ex. routeur) pour faire le stockage et le traitement des données. Cependant, le Fog présente une infrastructure hétérogène. En effet, ses équipements présentent des différences de performances de calcul, de capacités de stockage et d’interconnexions réseaux.Cette hétérogénéité peut davantage augmenter la latence du service. Cela pose un problème : le mauvais choix des emplacements de stockage des données peut augmenter la latence du service. Dans cette thèse, nous proposons une solution à ce problème sous la forme de quatre contributions : 1. Une formulation du problème de placement de données de l’IoT dans le Fog comme un programme linéaire. 2. Une solution exacte pour résoudre le problème de placement de données en utilisant CPLEX, un solveur de problème linéaire. 3. Deux heuristiques basées sur le principe de “diviser pour régner” afin de réduire le temps du calcul de placement. 4. Une plate-forme expérimentale pour évaluer des solutions de placement de données de l’IoT dans le Fog, en intégrant la gestion du placement de données à iFogSim, un simulateur d’environnement Fog et IoT. / In the coming years, Internet of Things (IoT) will be one of the applications generating the most data. Nowadays, IoT data is stored in the Cloud. As the number of connected objects increases, transmitting the large amount of produced data to the Cloud will create bottlenecks. As a result, latencies will be high and unpredictable. In order to reduce these latencies, Fog computing has been proposed as a paradigm extending Cloud services to the edge of the network. It consists of using any equipment located in the network (e.g. router) to store and process data. Therefore, the Fog presents a heterogeneous infrastructure. Indeed, its components have differences in computing performance, storage capacity and network interconnections. This heterogeneity can further increase the latency of the service. This raises a problem: the wrong choice of data storage locations can increase the latency of the service. In this thesis, we propose a solution to this problem in the form of four contributions: 1. A formulation of the IoT data placement problem in the Fog as a linear program. 2. An exact solution to solve the data placement problem using the CPLEX, a mixed linear problem solver. 3. Two heuristics based on the principle of “divide and conquer” to reduce the time of placement computation. 4. An experimental platform for testing and evaluating solutions for IoT data placement in the Fog, integrating data placement management with iFogSim, a Fog and IoT environment simulator.

Entre outubro de 2009 e dezembro de 2012, foi realizado estudo no Laboratório de Hidrologia Florestal Walter Emmerich (L.H.F.W.E.), para avaliar o potencial hidrológico dos nevoeiros e a precipitação oculta em ambiente de Floresta Ombrófila Densa Montana, na Serra do Mar, em Cunha, a leste do Estado de São Paulo. Os objetivos específicos da pesquisa foram: 1) avaliar o desempenho de coletores de água de nevoeiro; 2) verificar o potencial de coleta de água de nevoeiro no laboratório; 3) analisar o potencial de coleta de água de nevoeiro em três setores da microbacia experimental D; 4) averiguar a distribuição espacial e temporal de ocorrência dos nevoeiros na microbacia D e 5) estimar a contribuição da precipitação oculta na floresta da microbacia D. Para alcançar os três primeiros objetivos foram usados coletores passivos tipo harpa, nas formas cilíndrica e plana. Para estudar a distribuição de nevoeiros, foram feitas observações diárias de três referências visuais situadas a distâncias conhecidas no interior da microbacia D. O método da medição da precipitação efetiva foi empregado para estimar a precipitação oculta na floresta. Sobre a distribuição espacial e temporal da ocorrência de nevoeiros na microbacia D, os resultados mostraram que os episódios de nevoeiro foram mais comuns à tarde, como decorrência da brisa marítima proveniente do Oceano Atlântico. Para 61,3% do tempo, em média, foi observada a presença de nevoeiro em algum setor da microbacia, com a ocorrência dos nevoeiros diminuindo de montante para jusante da área. No que concerne ao desempenho dos equipamentos, a média de coleta diferiu de coletor para coletor, exceto entre o coletor cilíndrico descoberto e o coletor plano, que interceptaram mais água que os coletores cilíndricos cobertos. O coletor plano foi o único a interceptar água em todos os episódios amostrados. Sobre o potencial de coleta de água de nevoeiros no laboratório, concluiu-se que é reduzido e resulta da combinação de três fatores: a grande ocorrência de nevoeiros de radiação, a curta duração dos ix episódios dos nevoeiros orográficos e a baixa intensidade dos ventos na região. No que concerne ao potencial de coleta de água de nevoeiro em três setores da microbacia D, árvores localizadas próximas aos coletores funcionaram como obstáculo à livre circulação dos nevoeiros orográficos, induzindo os coletores a interceptar exíguos volumes de água. Considerando apenas os eventos de nevoeiro sem chuva, a precipitação oculta foi de 2,3 mm, correspondendo a 0,12% da precipitação pluviométrica anual. Para os eventos de chuva com a presença de nevoeiro, a precipitação oculta gerou um total de 5,1 mm de água adicional ao piso da floresta, correspondendo a 0,27% da precipitação pluviométrica. Concluiu-se que a precipitação oculta na floresta não é um processo importante para a entrada de água na microbacia D. / Between October 2009 and December 2012 a study was conducted at the Walter Emmerich Hydrologic Laboratory (LHFWE) to assess the hydrological potential of fog and occult precipitation in the Montane Dense Ombrophilous Forest environment located in the Parque Estadual da Serra do Mar - Núcleo Cunha, east of the State of São Paulo, Brazil. The specific multiple objectives of the research were: 1) evaluate the performance of passive fog collectors, 2) evaluate the potential for collecting fog water in the laboratory, 3) evaluate the potential of collecting fog water in three sectors of experimental catchment \"D\", 4) determine the spatial and temporal distributions of fog occurrence in catchment \"D\", 5) estimate the contribution of occult precipitation in the rainforest in catchment \"D\". To achieve the first three objectives, cylindrical and flat harp collectors were used. To study the spatial and temporal distributions of fog occurrence, we made daily observations of three visual references located at known distances within the catchment. Net precipitation was measured and used to estimate the occult precipitation in the rainforest. Regarding the spatial and temporal distributions of fog occurrence, the results showed that episodes of fog were more common in the afternoon as a result of the sea breeze from the Atlantic Ocean. For 61.3% of the time on average, we observed the presence of fog in some sector of the catchment, with its occurrence decreasing in areas located from upstream to downstream. Concerning the comparison of the performance of the collectors, the average volume of water collected differed from collector to collector, except between cylindrical collector uncovered and flat collector, which captured more water than the cylindrical collectors covered to prevent rainwater input. The flat collector was the only gauge to intercept water in all episodes. The high incidence of fog radiation, the short duration of orographic fog, and the low intensity of winds combined to reduce the potential for harvesting fog water in the laboratory. Regarding the potential for collecting fog water in three sectors of the catchment \"D\", xi trees located next to collectors operated as an obstacle to the free movement of orographic fog inducing gauges to intercept exiguous volumes of water. Considering only fog, but no rain, occult precipitation was 2.3 mm, corresponding to 0.12% of annual rainfall. For rainfall events with the presence of fog, occult precipitation generated a total of 5.1 mm of additional water to the floor of the forest, equivalent to 0.27% of the rainfall. It was therefore concluded that occult precipitation in rainforest did not significantly contribute to water entering catchment \"D\".

Cette thèse porte sur le Web Sémantique des Objets (WSdO), un domaine de recherche à l'interface de l'Internet des Objets (IdO) et du Web Sémantique (WS). L’intégration des approche du WS à l'IdO permettent de traiter l'importante hétérogénéité des ressources, des technologies et des applications de l'IdO, laquelle est une source de problèmes d'interopérabilité freinant le déploiement de systèmes IdO. Un premier verrou scientifique est lié à la consommation en ressource des technologies du WS, là où l'IdO s’appuie sur des objets aux capacités de calcul et de communication limitées. De plus, les réseaux IdO sont déployés à grande échelle, quand la montée en charge est difficile pour les technologies du WS. Cette thèse a pour objectif de traiter ce double défi, et comporte deux contributions. La première porte sur l'identification de critères de qualité pour les ontologies de l'IdO, et l’élaboration de IoT-O, une ontologie modulaire pour l'IdO. IoT-O a été implantée pour enrichir les données d'un bâtiment instrumenté, et pour être moteur de semIoTics, notre application de gestion autonomique. La seconde contribution est EDR (Emergent Distributed Reasoning), une approche générique pour distribuer dynamiquement le raisonnement à base de règles. Les règles sont propagées de proche en proche en s'appuyant sur les descriptions échangées entre noeuds. EDR est évaluée dans deux scénario concrets, s'appuyant sur un serveur et des noeuds contraints pour simuler le déploiement. / This thesis is situated in the Semantic Web of things (SWoT) domain, at the interface between the Internet of Things (IoT) and the Semantic Web (SW). The integration of SW approaches into the IoT aim at tackling the important heterogeneity of resources, technologies and applications in the IoT, which creates interoperability issues impeding the deployment of IoT systems. A first scientific challenge is risen by the resource consumption of the SW technologies, inadequated to the limites computation and communication capabilities of IoT devices. Moreover, IoT networks are deployed at a large scale, when SW technologies have scalability issues. This thesis addresses this double challenge by two contributions. The first one is the identification of quality criteria for IoT ontologies, leading to the proposition of IoT-O, a modular IoT ontology. IoT-O is deployed to enrich data from a smart building, and drive semIoTics, our autonomic computing application. The second contribution is EDR (Emergent Distributed Reasoning), a generic approach to dynamically distributed rule-based reasoning. Rules are propagated peer-to-peer, guided by descriptions exchanged among nodes. EDR is evaluated in two use-cases, using both a server and some constrained nodes to simulate the deployment.

Le brouillard cause des dangers pour le trafic par la réduction de visibilité. L’amélioration des prévisions du brouillard est donc un objectif scientifique. Cette thèse analyse le cycle de vie des brouillards continentaux autour de Paris, observés par télédétection au sol à l’observatoire atmosphérique SIRTA. La thèse se focalise sur la compréhension des processus en jeu dans la dissipation après le lever du soleil, sous l’hypothèse d’une couche de brouillard adiabatique. Pendant 4 ans, plus de 100 événement de brouillard sont documentés par l’observation de la base du nuage (par télémètre), son sommet et la présence de nuages au-dessus (radar nuage), et le contenu intégré d’eau liquide (LWP) (radiomètre micro-onde (MWR)). La plupart des brouillards se dissipe suite à un soulèvement de la base, sans que tout le nuage s’évapore, et souvent sans une réduction du LWP. Donc, non seulement est la réduction du LWP importante pour la dissipation du brouillard, mais aussi l’évolution de son sommet, qui avec le LWP détermine l’altitude de la base. Des simulations par le modèle LES DALES montrent une sensibilité importante à la stratification au-dessus : en augmentant l’entrainement, une stratification faible au sommet peut accélérer la dissipation par (1) plus de perte d’eau liquide par l’entrainement de l’air non-saturé, et (2) par un développement vertical menant au lever de la base. La variabilité de cette stratification peut être raisonnablement bien observée par le profil de température du MWR. Avant la dissipation du brouillard par lever de la base, le radar observe souvent un max de réflectivité près du sommet, ce qui peut être lié à l’absence de grandes gouttelettes dans les basses couches. Donc, par leur observation du développement du sommet, le LWP, la stratification, et le profil de réflectivité, le radar et le MWR donnent des informations qui peuvent potentiellement anticiper la dissipation du brouillard.Les processus radiatifs sont étudiés avec le code de transfert radiatif ARTDECO. Le refroidissement radiatif au sommet du brouillard peut produire 40–70 g m-2 h-1 d’LWP quand le brouillard est opaque (LWP >= 30 g m-2) (c’est moins pour les brouillards minces) et il n’y a pas de nuage au-dessus. C’est la source principale d’LWP et il peut renouveler le LWP du brouillard en 0.5–2 h. Sa variabilité s’explique principalement par la température du brouillard et le profil d’humidité au-dessus. Les nuages au-dessus du brouillard réduisent fortement la production, en particulier les nuages bas. La perte d’LWP par absorption de rayonnement solaire par le brouillard est 5–15 g m-2 h-1 autour de midi en hiver, dépendant de l’épaisseur du brouillard, mais ça peut augmenter par 100 % quand une quantité importante d’aérosols absorbants est présente (AOD=0.15, SSA=0.82).Nos résultats par simulation LES indiquent que le réchauffement par absorption de rayonnement solaire à la surface est le premier processus de perte d’LWP après le lever du soleil, mais sa magnitude est sensible au rapport de Bowen. Vu son importance, une amélioration de l’observation du rapport de Bowen dans le brouillard devrait être une priorité, car les observations actuelles des flux turbulents ne sont pas suffisamment précises pour quantifier le rapport de Bowen.Un modèle conceptuel pour calculer le bilan du LWP directement à partir des observations est développé. En utilisant 12 paramètres observés et 2 qui viennent d’une réanalyse, il calcule les impacts au LWP par rayonnement, flux de chaleur à la surface, entrainement, subsidence et dépôt. Ce modèle est appliqué à 45 brouillards observés qui se dissipent après le lever du soleil. Une variabilité importante dans le rayonnement, l’entrainement et la subsidence entre les cas est trouvée, qui peut en partie expliquer les différences en heure de dissipation. Tandis que les termes de rayonnement sont plutôt précis, des autres ont des incertitudes importantes et pourront être améliorés dans le futur. / Fog causes hazards to human activity due to the reduction of visibility, especially through the risk of traffic accidents. Improving the forecasts of fog formation and dissipation is therefore an objective for research. This thesis analyses the life cycle of continental fog events occurring in the Paris area, using several ground-based remote sensing instruments deployed at the SIRTA atmospheric observatory. We focus on understanding the dissipation after sunrise and the local processes involved, assuming the fog layer is adiabatic (well-mixed). Over a 4-year period, more than 100 fog events are documented by observing cloud base (ceilometer), cloud top and clouds appearing above the fog (cloud radar), and the liquid water path (LWP) (microwave radiometer (MWR)). Most fog events dissipate by lifting of the base without a complete evaporation of the cloud, and often even without a reduction in LWP. This indicates that not only a reduction in LWP is important for fog dissipation, but also the evolution of the fog top, which together with the LWP determines whether the cloud extends down to the ground. Using the LES model DALES, we find a strong sensitivity of the vertical development of the fog top to the stratification above. By enhancing entrainment, a weak stratification at fog top can lead to earlier fog dissipation by (1) more depletion of LWP by entraining unsaturated air, especially if the air is dry, and (2) vertical development of the fog top leading to lifting of the fog base. The variability of this stratification can be observed reasonably well with the MWR temperature profile. In several cases of dissipation by lifting, the vertical profile of radar reflectivity in the fog has a max value near fog top prior to dissipation, which suggests a lack of bigger droplets in the lower levels of the fog. By observing the cloud top development, the stratification, the LWP and the profile of reflectivity, the radar and MWR provide information that has potential for anticipating fog dissipation by lifting.Radiative processes are studied using the comprehensive radiative transfer code ARTDECO. The radiative cooling at fog top can produce 40–70 g m-2 h-1 of LWP when the fog is opaque (LWP >= 30 g m-2) (production is lower for thin fog) and there are no clouds above. This cooling thus is the main process of LWP production and can renew the fog LWP in 0.5–2 h. Its variability is mainly explained by the fog temperature and the humidity profile above. Clouds above the fog will strongly reduce this production, especially low clouds: a cloud with optical depth 4 can reduce it by 30 (100) % at 10 (2) km. Loss of LWP by absorption of solar radiation by the fog is 5–15 g m-2 h-1 around midday in winter, depending on cloud thickness, but it can be enhanced by 100 % in case of important amounts of absorbing aerosols (dry AOD=0.15, SSA=0.82).Heating due to solar radiation absorbed at the surface is found to be the dominating process of LWP loss after sunrise (according to LES model simulations), but its magnitude is sensitive to the Bowen ratio. However, observations of the turbulent heat fluxes during fog are not precise enough to quantify the Bowen ratio. The importance of the Bowen ratio means that improvements of its measurement during fog should be a priority.A conceptual model which calculates the LWP budget of fog directly from observations is developed. Using 12 observed parameters and 2 from reanalysis data, it calculates the impact on LWP of terrestrial and solar radiation, surface heat fluxes, entrainment, subsidence and deposition. It is applied to 45 observed fog events dissipating after sunrise. An important variability in radiation, entrainment and subsidence between the cases is found, which can partly explain the different dissipation times. While the terms of radiation are rather robust, several other terms suffer from significant uncertainties, leaving room for improvements in the future.

L'informatique pervasive promeut une vision d'un cadre dans lequel un patchwork de ressources hétérogènes et volatiles est intégré dans les environnements du quotidien. Ces ressources, matérielles ou logicielles, coopèrent de manière transparente, souvent aux travers d'applications, pour fournir des services à haute valeur ajoutée adaptés à chaque utilisateur et son environnement, grâce à la notion de contexte. Ces applications sont déployées dans un large spectre d'environnements d'exécution, allant d'infrastructures distantes de Cloud Computing jusqu'au plus près de l'utilisateur dans des passerelles Fog Computing ou directement dans les capteurs du réseau. Dans ces travaux, nous nous intéressons spécifiquement au module de contexte d'une plateforme Fog Computing. Pour faciliter la conception et l'exécution des applications Fog Computing, une approche populaire est de les bâtir au dessus d'une plateforme adoptant l'architecture à service, ce qui permet de réduire leur complexité et simplifie la gestion du dynamisme. Dans nos travaux, nous proposons d'étendre cette approche en modélisant le contexte comme un ensemble de descriptions de services, disponible à la conception, et exposé dynamiquement par le module de contexte à l'exécution, selon les besoins des applications et l'état de l'environnement. Ce module est programmé à l'aide d'un modèle à composant spécifique. L'unité de base de notre modèle à composant est l'entité de contexte, qui est composé de modules hautement cohérents implémentant distinctement les spécifications des services proposées par l'entité de contexte. Ces modules peuvent décrire de manière simple leur logique de synchronisation avec les sources de contexte distantes grâce à un langage dédié à ce domaine. A l'exécution, les instances d'entitées de contexte sont rendues introspectables et reconfigurables dynamiquement, ce qui permet, grâce à un manager autonomique externe, de veiller à la satisfaction des besoins des applications. Nous avons développé une implémentation de référence de ce modèle à composant, nommée CReAM, qui a pu être utilisée dans la passerelle domotique iCASA, développée en partenariat avec Orange Labs. / Pervasive computing promotes environments where a patchwork of heterogeneous and volatile resources are integrated in places of daily life. These hardware and software resources cooperate in a transparent way, through applications, in order to provide high valueadded services. These services are adapted to each user and its environment, via the notion of context. Pervasive applications are now widely distributed, from distant cloud facilities down to Fog Computing gateway or even in sensors, near the user. Depending on the localization, various forms of context are needed by the applications. In this thesis, we focus on the context module at Fog Level. In order to simplify the design and execution, Fog applications are built on top of a service-oriented platform, freeing the developer of technical complexity and providing a support to handle the dynamism. We propose to extend this approach by providing the context as a set of service descriptions, available at design to the application developer. At runtime, depending on the context sources availability and on application current needs, context services are published or withdrawn inside the platform by the context module. We tailor a specific component model to program this context module. The base unit of our component model is called context entity. It is composed of highly coherent modules, implementing distinctly each service description proposed by the underlying context entity. These modules can simply describe their synchronization logic with context sources thanks to a domain specific language. At runtime, context entity instances can be introspected and reconfigured. An external autonomic manager uses these properties to match dynamically the context services exposed by the context module to the application needs. We have developed a reference implementation of our work, called CReAM, which can be used in a smart home gateway called iCASA, developed in a partnership with Orange Labs.

Made available in DSpace on 2014-12-17T14:57:55Z (GMT). No. of bitstreams: 1 jailtonWG.pdf: 3426096 bytes, checksum: a6ff181ad34e9b5be394de7ebee923b3 (MD5) Previous issue date: 2009-04-24 / An alternative box-type solar cooker built starting from the scrap of a tire and a scrap of old office chair is presented, which principles functions are the effect greenhouse and the concentration. The tire served as structure for making of is the baking enclosure where the absorber (roasting pan 20x30cm) of the solar is located, being re-covered for a glass blade for the generation of the greenhouse effect isolated lateral and having deep its and for a composite the plaster base and EPS. Segments of plain mirrors had been placed in the laterals of the oven/cook for the concentration of the radiation and a reflecting parable was introduced in the baking enclosure for the exploitation of the incident reflected radiation inside of the oven/cook. The oven/cook is mobile to allow one better aiming of exactly in relation to the apparent movement of the sun. The thermal economic and of materials viabilities of the stove/cook in study will be demonstrate. The average internal temperature of the absorber was around 152,3?C and the internal temperature around 110?C. Will demonstrate that toits low cost and good thermal performance, represents basic characteristics for the viability of large use of such archetype, mainly for cooking the decreases and averages temperatures. One will reveal that the archetype in study is competitive with the box-type solar cooker conceived in the whole world / Apresenta-se um forno/fog?o solar alternativo de baixo custo para ser utilizado nas opera??es de assar e cozinhar, que foi constru?do a partir de uma sucata de pneu. O pneu foi cortado na parte superior para aumentar a ?rea de capta??o da radia??o solar incidente. Utilizou- se comp?sito a base de gesso, EPS (poliestireno expandido) triturado e ?gua para revestir internamente a carca?a do pneu e o fundo da mesma para minimizar as perdas t?rmicas. Ser?o estudadas tr?s configura??es, onde ser?o utilizados dois tipos de par?bolas refletoras. Uma confeccionada a partir de uma tampa de prote??o de uma sucata de ventilador e outra a partir de uma urupema, peneira usada para comidas de milho. Acima da cobertura de vidro foi colocada uma estrutura com espelhos planos para refletir os raios incidentes para o interior do forno/fog?o solar. A estrutura de sustenta??o do forno/fog?o solar, com os movimentos necess?rios ao acompanhamento do movimento aparente do sol foi confeccionada utilizando uma sucata de cadeira girat?ria. O forno/fog?o solar proposto apresenta maior viabilidade para a opera??o de assar. Obteve-se uma temperatura m?xima no absorvedor em torno de 160? e interna em torno de 120?C. Demonstrar-se-? as viabilidades t?rmica, econ?mica e de materiais do forno/fog?o solar proposto

Dag för dag blir sakernas internet-enheter (IoT) en större del av vårt liv. För närvarande är dessa enheter starkt beroende av molntjänster vilket kan utgöra en integritetsrisk. Det allmänna syftet med denna rapport är att undersöka alternativ till molntjänster, ett ganska fascinerande alternativ är fog computing. Fog computing är en struktur som utnyttjar processorkraften hos enheter i utkanten av nätverket (lokala enheter) snarare än att helt förlita sig på molntjänster. Ett specifikt fall av denna struktur undersöks ytterligare som huvudsyftet i denna rapport vilket är distribuerad maskininlärning för IoT-enheter. Detta mål uppnås genom att besvara frågorna om vilka metoder/verktyg som finns tillgängliga för att åstadkomma det och hur väl fungerar de jämfört med molntjänster. Det finns tre huvudsteg i denna studie. Det första steget var informationsinsamling på två olika nivåer. Först på en grundläggande nivå där området för studien undersöks. Den andra nivån var mer specifik och handlade om att ytterligare samla information om tillgängliga verktyg för distribuering av maskininlärning och utvärdera dessa verktyg. Det andra steget var att implementera tester för att verifiera prestandan för varje verktyg vald baserat på den insamlade informationen. Det sista steget var att sammanfatta resultaten och dra slutsatser. Studien har visat att distribuerad maskininlärning fortfarande är för omogen för att ersätta molntjänster eftersom de befintliga verktygen inte är optimerade för IoT-enheter. Det bästa alternativet för tillfället är att hålla sig till molntjänster, men om lägre prestanda till viss del kan tolereras, så är vissa IoT-enheter kraftfulla nog att bearbeta maskininlärningsuppgiften självständigt. Distribuerad maskininlärning är fortfarande ett ganska nytt koncept, men det utvecklas snabbt, förhoppningsvis når denna utveckling snart IoT-enheter. / By day, internet of things (IoT) devices is becoming a bigger part of our life. Currently these devices are heavily dependent on cloud computing which can be a privacy risk. The general aim of this report is to investigate alternatives to cloud computing, a quite fascinating alternative is fog computing. Fog computing is a structure that utilizes the processing power of devices at the edge of the network (local devices) rather than fully relying on cloud computing. A specific case of this structure is further investigated as the main objective of this report which is distributed machine learning for IoT devices. This objective is achieved by answering the questions of what methods/tools are available to accomplish that and how well do they function in comparison to cloud computing. There are three main stages of this study. The first stage was information gathering on two different levels. First on a basic level exploring the field. The second one was to further gather information about available tools for distributing machine learning and evaluate them. The second stage was implementing tests to verify the performance of each approach/tool chosen from the information gathered. The last stage was to summarize the results and reach to conclusions. The study has shown that distributed machine learning is still too immature to replace cloud computing since the existing tools isn’t optimized for this use case. The best option for now is to stick to cloud computing, but if lower performance to some extent can be tolerated, then some IoT devices is powerful enough to process the machine learning task independently. Distributed machine learning is still quite a new concept but it’s growing fast, hoping this growth soon expands to support IoT devices.

Récemment, les villes sont devenues "de plus en plus intelligentes", avec une multitude de périphériques IoT et de capteurs déployés partout. Parmi ces objets intelligents, les voitures peuvent jouer un rôle important. Les véhicules sont (ou seront), en effet, équipés avec plusieurs interfaces réseau, ils ont (ou auront) des capacités de calcul et des dispositifs capables d'analyser l'environnement. Pour réaliser le concept de "connected-car" il faut un changement de modèle Internet, à partir d'une architecture centrée sur l'hôte (IP) vers un paradigme centré sur l'information, comment l'architecture ICN (Information Centric Networking). Cette thèse analyse ainsi les avantages et les défis du paradigme ICN, en particulier du Named Data Networking (NDN), dans le domaine VANET, en présentant la première implémentation de NDN pour VANET (V-NDN). Il propose ensuite Navigo, un mécanisme de forwarding basé sur NDN pour la récupération de contenu en utilisant les communications V2V et V2I. Ensuite, le problème de la mobilité des fournisseurs de données est traité, proposant une solution distribuée basée sur NDN, MAP-Me. Toutefois, le rôle du véhicule dans les villes intelligentes ne s'arrête pas au niveau de la connectivité. Les voitures, avec leurs nouvelles capacités de calcul, sont les candidates idéales pour jouer un rôle dans l'architecture Fog Computing, en déplaçant des tâches de calcul vers l'edge du réseau. En tant que preuve de concept, cette thèse présente ParkMaster, un système qui combine les techniques de machine learning, le cloud et l'edge pour analyser l'environnement et traiter le problème de la disponibilité du stationnement. / In the recent years we have seen a continuous integration of technology with the urban environment. This fusion aims to improve the efficiency and the quality of living in big urban agglomerates, while reducing the costs for their management. Cities are getting “smarter and smarter”, with a plethora of IoT devices and sensors deployed all over the urban areas. Among those intelligent objects, an important role may be played by cars. Modern vehicles are (or will be) indeed equipped with multiple network interfaces, they have (or will have) computational capabilities and devices able to sense the environment. However, smart and connected cars do not represent only an opportunity, but also a challenge. Computation capabilities are limited, mobility and the diversity of network interfaces are obstacles when providing connectivity to the Internet and to other vehicles. When addressing the networking aspect, we believe that a shift in the Internet model is needed, from a host oriented architecture (IP) to a more content focused paradigm, the Information Centric Networking (ICN) architectures. This thesis thus analyzes the benefits and the challenges of the ICN paradigm, in particular of Named Data Networking (NDN), in the VANET domain, presenting the first implementation running on real cars of NDN for VANET (V-NDN). It then proposes Navigo, an NDN based forwarding mechanism for content retrieval over V2V and V2I communications, with the goal of efficiently discovering and retrieving data while reducing the network overhead. Networking mobility is not only a challenge for vehicles, but for any connected mobile device. For this reason, this thesis extends its initial area of interest — VANET — and addresses the network mobility problem for generic mobile nodes, proposing a NDN-based solution, dubbed MAP–Me. MAP-Me tackles the intra-AS content provider mobility problem without relying on any fixed node in the network. It exploits notifications messages at the time of a handover and the forwarding plane to maintain the data provider “always” reachable.Finally, the “connected car” concept is not the only novel element in modern vehicles. Cars indeed won’t be only connected, but also smart, able to locally process data produced by in-car sensors. Vehicles are the perfect candidates to play an important role in the recently proposed Fog Computing architecture. Such an architecture moves computational tasks typical of the cloud away from it and brings them to the edge, closer to where the data is produced. To prove that such a model, with the car as computing edge node, is already feasible with the current technology and not only a vision for the future, this thesis presents ParkMaster. Parkmaster is a fully deployed edge-based system that combines vision and machine learning techniques, the edge (driver’s smartphone) and the cloud to sense the environment and tackle the parking availability problem.

La technologie et les conducteurs socio-économiques créent les conditions d'une transformation profonde, appelée "Softwarization", du Telco et des TIC. Réseaux définis par logiciel et réseau Fonctions de virtualisation sont deux des principales technologies permettant ouvrant la voie à cette transformation. Softwarization permettra de virtualiser toutes les fonctions de réseau et de services d'une infrastructure de Telco et de les exécuter sur une plates-formes logicielles, entièrement découplés de l'infrastructure physique sous (presque basé sur du matériel standard). Tous les services seront fournis en utilisant un «continuum» des ressources virtuelles (traitement, de stockage et de communication) avec un investissement en capital initial pratiquement très limité et avec des coûts d'exploitation modestes. 5G sera la première exploitation de Softwarization. 5G sera une infrastructure distribuée massivement dense, intégrant le traitement, le stockage et (fixes et radio) des capacités de mise en réseau. En résumé, l'objectif général de cette thèse a étudié les défis techniques et les opportunités d'affaires apportées par le "Softwarization" et 5G. En particulier, la thèse propose que le 5G devra avoir une sorte de système d'exploitation (5GOS) capable de fonctionner les RAN et de base et les infrastructures fixes convergés. Les contributions de cette thèse ont été: 1) définir une vision pour les futures infrastructures 5G, des scénarios, des cas d'utilisation et les exigences principales: 2) définissant l'architecture fonctionnelle d'un système d'exploitation pour 5G; 3) la conception de l'architecture logicielle d'un 5GOS pour le "bord Cloud"; 4) comprendre les impacts technico-économiques de la vision et 5GOS, et les stratégies les plus efficaces pour l'exploiter / Technology and socio-economic drivers are creating the conditions for a profound transformation, called “Softwarization”, of the Telco and ICT. Software-Defined Networks and Network Functions Virtualization are two of the key enabling technologies paving the way towards this transformation. Softwarization will allow to virtualize all network and services functions of a Telco infrastructure and executing them onto a software platforms, fully decoupled from the underneath physical infrastructure (almost based on standard hardware). Any services will be provided by using a “continuum” of virtual resources (processing, storage and communications) with practically very limited upfront capital investment and with modest operating costs. 5G will be the first exploitation of Softwarization. 5G will be a massively dense distributed infrastructure, integrating processing, storage and (fixed and radio) networking capabilities. In summary, the overall goal of this thesis has been investigating technical challenges and business opportunities brought by the “Softwarization” and 5G. In particular, the thesis proposes that the 5G will have to have a sort of Operating System (5GOS) capable of operating the converged fixed and RAN and core infrastructures. Main contributions of this thesis have been: 1) defining a vision for future 5G infrastructures, scenarios, use-cases and main requirements; 2) defining the functional architecture of an Operating System for 5G; 3) designing the software architecture of a 5G OS for the “Edge Cloud”; 4) understanding the techno-economic impacts of the vision and 5GOS, and the most effective strategies to exploit it

The Internet of Things is a recent computing paradigm, de- fined by networks of highly connected things – sensors, actuators and smart objects – communicating across networks of homes, buildings, vehicles, and even people. The Internet of Things brings with it a host of new problems, from managing security on constrained devices to processing never before seen amounts of data. While cloud computing might be able to keep up with current data processing and computational demands, it is unclear whether it can be extended to the requirements brought forth by Internet of Things. Fog computing provides an architectural solution to address some of these problems by providing a layer of intermediary nodes within what is called an edge network, separating the local object networks and the Cloud. These edge nodes provide interoperability, real-time interaction, routing, and, if necessary, computational delegation to the Cloud. This paper attempts to evaluate Go, a distributed systems language developed by Google, in the context of requirements set forth by Fog computing. Similar methodologies of previous literature are simulated and benchmarked against in order to assess the viability of Go in the edge nodes of Fog computing architecture.