Thèses sur le sujet « Traces distribuées »

L'objectif de cette thèse est d'établir certaines bases théoriques, méthodologiques et pratiques d'une " sociologie assistée par ordinateur " des pratiques collectives distribuées (PCD). Le développement des nouvelles technologies de l'information et de la communication a mené à l'émergence de formes d'organisation dont les principales caractéristiques sont la forte distribution (spatiale et temporelle, mais aussi socio-cognitive) et l'utilisation prépondérante de moyens de communication médiés, qui laissent des " traces " persistantes de l'activité collective. Nos travaux portent sur l'exploitation de ces traces, et en particulier des traces d'interactions en langue naturelle, en tant que moyen d'analyse de l'activité sous-jacente du collectif, afin de mieux caractériser " ce qui se passe " dans le collectif. Les très grandes quantités de données disponibles nous poussent à chercher à développer des méthodes d'analyse automatisées capables de traiter les contenus langagiers de telles traces. Notre approche a consisté à 1) concevoir un modèle capable de représenter les interactions d'un collectif distribué et leurs relations avec l'activité du collectif, 2) évaluer l'utilité d'un tel modèle pour l'étude des PCD au travers d'une phase expérimentale et 3) étudier la faisabilité de l'automatisation des traitements requis par le modèle à l'aide de technologies d'apprentissage machine et de traitement du langage. Notre étude a plus particulièrement porté sur des données recueillies dans le collectif open-source Bugzilla
The aim of this thesis is to establish some of the theoretical, methodological and practical foundations of a “computer-supported sociology” of distributed collective practices (DCP). The development of new information and communication technologies lead to the emergence of organization forms which main characteristics are their large-scale distribution (spatial and temporal, but also socio-cognitive) and the central use of mediated communication channels, which leave persistent “traces” of collective activity. Our work focuses on exploiting these traces, especially traces of natural language interaction, as a mean of studying the underlying activity of the collective, in order to better study “what is going on” in the collective. The large amounts of data available lead us to attempt to elaborate methods relying on automated analyses and capable of handling the linguistic content of these traces. Our approach consisted in 1) designing a model capable of representing distributed collective interaction and the relations between such interactions and the collective's activity, 2) evaluating the usefulness of such model for the study of DCP through an experimental phase, and 3) studying the feasibility of automating the data processing needed by the model through use of machine learning and language processing technologies. Our study focused on data collected in the Bugzilla open-source collective

Le Cloud Computing a bouleversé la façon dont sont développés et déployées les logiciels. De nos jours, les applications Cloud sont conçues comme des systèmes distribués, en constante évolution, hébergés dans des data~center gérés par des tiers, et potentiellement même dispersés dans le monde entier. Ce changement de paradigme a également eu un impact considérable sur la façon dont les logiciels sont monitorés : les applications cloud se sont développées pour atteindre l'ordre de centaines de services, et les outils de monitoring ont rapidement rencontré des problèmes de mise à l'échelle. De plus, ces outils de monitoring doivent désormais également traiter les défaillances et les pannes inhérentes aux systèmes distribués, comme par exemple, les pannes partielles, les configurations incohérentes, les goulots d'étranglement ou même la vampirisation de ressources. Dans cette thèse, nous présentons une approche basée sur une nouvelle source de télémétrie qui s'est développée dans le domaine du monitoring des applications Cloud. En effet, en nous appuyant sur le récent standard OpenTelemetry, nous présentons un système qui convertit les données de "traces distribuées" en un graphe de propriétés hiérarchique. Grâce un tel modèle, il devient possible de mettre en évidence la topologie des applications, y compris la répartition sur les différentes machines des programmes, y compris sur plusieurs data-centers. L'objectif de ce modèle est donc d'exposer le comportement des fournisseurs de service Cloud aux développeurs qui maintiennent et optimisent leur application. Ensuite, nous présentons l'utilisation de ce modèle pour résoudre certains des défis majeurs des systèmes distribués~: la détection des communications inefficaces entre les services et l'anticipation des goulots d'étranglement. Nous abordons ces deux problèmes avec une approche basée sur la théorie des graphes. La composition inefficace des services est détectée avec le calcul de l'indice de hiérarchie de flux. Une plateforme Proof-of-Concept représentant un cluster Kubernetes zonal pourvu d'une instrumentation OpenTelemetry est utilisée pour créer et détecter les compositions de services inefficaces. Dans une dernière partie, nous abordons la problématique de la détection des goulots d'étranglement dans un réseau de services au travers de l'analyse de centralité du graphe hiérarchique précédent. Ce travail s'appuie sur un programme de simulation qui a aussi été instrumenté avec OpenTelemetry afin d'émettre des données de traçage. Ces traces ont été converties en un graphe de propriétés hiérarchique et une étude sur les algorithmes de centralité a permis d'identifier les points d'étranglement. Les deux approches présentées dans cette thèse utilisent et exploitent l'état de l'art en matière de monitoring des applications Cloud. Elles proposent une nouvelle utilisation des données de "distributed tracing" pas uniquement pour l'investigation et le débogage, mais pour la détection et la réaction automatiques sur un système réel
Cloud Computing has changed how software is now developed and deployed. Nowadays, Cloud applications are designed as rapidly evolving distributed systems that are hosted in third-party data centre and potentially scattered around the globe. This shift of paradigms also had a considerable impact on how software is monitored: Cloud application have been growing to reach the scale of hundreds of services, and state-of-the-art monitoring quickly faced scaling issues. In addition, monitoring tools also now have to address distributed systems failures, like partial failures, configuration inconsistencies, networking bottlenecks or even noisy neighbours. In this thesis we present an approach based on a new source of telemetry that has been growing in the realm of Cloud application monitoring. Indeed, by leveraging the recent OpenTelemetry standard, we present a system that converts "distributed tracing" data in a hierarchical property graph. With such a model, it becomes possible to highlight the actual topology of Cloud applications like the physical distribution of its workloads in multiple data centres. The goal of this model is to exhibit the behaviour of Cloud Providers to the developers maintaining and optimizing their application. Then, we present how this model can be used to solve some prominent distributed systems challenges: the detection of inefficient communications and the anticipation of hot points in a network of services. We tackle both of these problems with a graph-theory approach. Inefficient composition of services is detected with the computation of the Flow Hierarchy index. A Proof of Concept is presented based on a real OpenTelemetry instrumentation of a Zonal Kubernetes Cluster. In, a last part we address the concern of hot point detection in a network of services through the perspective of graph centrality analysis. This work is supported by a simulation program that has been instrumented with OpenTelemetry in order to emit tracing data. These traces have been converted in a hierarchical property graph and a study on the centrality algorithms allowed to identify choke points. Both of the approaches presented in this thesis comply with state-of-the-art Cloud application monitoring. They propose a new usage of Distributed Tracing not only for investigation and debugging but for automatic detection and reaction on a full system

Dans cette thèse, nous étudions des algorithmes pour le monitoring des traces d’activité à grande échelle. Le monitoring est une aptitude clé dans plusieurs domaines, permettant d’extraire de la valeur des données ou d’améliorer les performances d’un système. Nous explorons d’abord le monitoring de données temporelles. Nous présentons un nouveau type de jointure sur des intervalles, qui inclut des fonctions de score caractérisant le degré de satisfaction de prédicats temporels. Nous étudions ces jointures dans le contexte du batch processing (traitement par lots). Nous formalisons la Ranked Temporal Join (RTJ), une jointure qui combine des collections d’intervalles et retourne les k meilleurs résultats. Nous montrons comment exploiter les propriétés des prédicats temporels et de la sémantique de score associée afin de concevoir TKIJ , une méthode d’évaluation de requête distribuée basée sur Map-Reduce. Nos expériences sur des données synthétiques et réelles montrent que TKIJ est plus performant que les techniques de l’état de l’art et démontre de bonnes performances sur des requêtes RTJ n-aires sur des données temporelles. Nous proposons également une étude préliminaire afin d’étendre nos travaux sur TKIJ au domaine du stream processing (traitement de flots). Nous explorons ensuite le monitoring dans le crowdsourcing (production participative). Nous soutenons la nécessité d’intégrer la motivation des travailleurs dans le processus d’affectation des tâches. Nous proposons d’étudier une approche adaptative, qui évalue la motivation des travailleurs lors de l’exécution des tâches et l’exploite afin d’améliorer l’affectation de tâches qui est réalisée de manière itérative. Nous explorons une première variante nommée Individual Task Assignment (Ita), dans laquelle les tâches sont affectées individuellement, un travailleur à la fois. Nous modélisons Ita et montrons que ce problème est NP-Difficile. Nous proposons trois méthodes d’affectation de tâches qui poursuivent différents objectifs. Nos expériences en ligne étudient l’impact de chaque méthode sur la performance globale dans l’exécution de tâches. Nous observons que différentes stratégies sont dominantes sur les différentes dimensions de performance. En particulier, la méthode affectant des tâches aléatoires et correspondant aux intérêts d’un travailleur donne le meilleur flux d’exécution de tâches. La méthode affectant des tâches correspondant au compromis d’un travailleur entre diversité et niveau de rémunération des tâches donne le meilleur niveau de qualité. Nos expériences confirment l’utilité d’une affectation de tâches adaptative et tenant compte de la motivation. Nous étudions une deuxième variante nommée Holistic Task Assignment (Hta), où les tâches sont affectées à tous les travailleurs disponibles, de manière holistique. Nous modélisons Hta et montrons que ce problème est NP-Difficile et MaxSNP-Difficile. Nous développons des algorithmes d’approximation pour Hta. Nous menons des expériences sur des données synthétiques pour évaluer l’efficacité de nos algorithmes. Nous conduisons également des expériences en ligne et comparons notre approche avec d’autres stratégies non adaptatives. Nous observons que notre approche présente le meilleur compromis sur les différentes dimensions de performance
In this thesis, we study scalable algorithms for monitoring activity traces. In several domains, monitoring is a key ability to extract value from data and improve a system. This thesis aims to design algorithms for monitoring two kinds of activity traces. First, we investigate temporal data monitoring. We introduce a new kind of interval join, that features scoring functions reflecting the degree of satisfaction of temporal predicates. We study these joins in the context of batch processing: we formalize Ranked Temporal Join (RTJ), that combine collections of intervals and return the k best results. We show how to exploit the nature of temporal predicates and the properties of their associated scored semantics to design TKIJ , an efficient query evaluation approach on a distributed Map-Reduce architecture. Our extensive experiments on synthetic and real datasets show that TKIJ outperforms state-of-the-art competitors and provides very good performance for n-ary RTJ queries on temporal data. We also propose a preliminary study to extend our work on TKIJ to stream processing. Second, we investigate monitoring in crowdsourcing. We advocate the need to incorporate motivation in task assignment. We propose to study an adaptive approach, that captures workers’ motivation during task completion and use it to revise task assignment accordingly across iterations. We study two variants of motivation-aware task assignment: Individual Task Assignment (Ita) and Holistic Task Assignment (Hta). First, we investigate Ita, where we assign tasks to workers individually, one worker at a time. We model Ita and show it is NP-Hard. We design three task assignment strategies that exploit various objectives. Our live experiments study the impact of each strategy on overall performance. We find that different strategies prevail for different performance dimensions. In particular, the strategy that assigns random and relevant tasks offers the best task throughput and the strategy that assigns tasks that best match a worker’s compromise between task diversity and task payment has the best outcome quality. Our experiments confirm the need for adaptive motivation-aware task assignment. Then, we study Hta, where we assign tasks to all available workers, holistically. We model Hta and show it is both NP-Hard and MaxSNP-Hard. We develop efficient approximation algorithms with provable guarantees. We conduct offline experiments to verify the efficiency of our algorithms. We also conduct online experiments with real workers and compare our approach with various non-adaptive assignment strategies. We find that our approach offers the best compromise between performance dimensions thereby assessing the need for adaptability

Le monitoring consiste a generer des informations de trace durant une execution d'un programme parallele pour detecter les problemes de performances. La quantite d'information generee par de tres grosses machines paralleles rend les outils d'analyse classiques inutilisables. Cette these resout ce probleme en distribuant l'information de trace sur plusieurs fichiers contenus sur plusieurs sites, les fichiers pouvant etre lus en parallele. La manipulation de ces fichiers afin d'obtenir une information coherente est la base d'un logiciel client-serveur grace auquel des clients demandent de l'information deja filtree sur une execution. Cette architecture client serveur est extensible (l'utilisateur peut creer ses propres clients) et modulable. Nous avons, d'autre part, cree deja plusieurs clients novateurs: client hierarchique, sonore, recherche automatique de problemes, interface filtrante aux outils classique, integration d'outil 3D

L'auteur n'a pas fourni de résumé en français
High Performance Computing is preparing the era of the transition from Petascale to Exascale. Distributed computing systems are already facing new scalability problems due to the increasing number of computing resources to manage. It is now necessary to study in depth these systems and comprehend their behaviors, strengths and weaknesses to better build the next generation.The complexity of managing users applications on the resources conducted to the analysis of the workload the platform has to support, this to provide them an efficient service.The need for workload comprehension has lead to the collection of traces from production systems and to the proposal of a standard workload format. These contributions enabled the study of numerous of these traces. This also lead to the construction of several models, based on the statistical analysis of the different workloads from the collection.Until recently, existing workload traces did not enabled researchers to study the consumption of resources by the jobs in a temporal way. This is now changing with the need for characterization of jobs consumption patterns.In the first part of this thesis we propose a study of existing workload traces. Then we contribute with an observation of cluster workloads with the consideration of the jobs resource consumptions over time. This highlights specific and unattended patterns in the usage of resources from users.Finally, we propose an extension of the former standard workload format that enables to add such temporal consumptions without loosing the benefit of the existing works.Experimental approaches based on workload models have also served the goal of distributed systems evaluation. Existing models describe the average behavior of observed systems.However, although the study of average behaviors is essential for the understanding of distributed systems, the study of critical cases and particular scenarios is also necessary. This study would give a more complete view and understanding of the performance of the resources and jobs management. In the second part of this thesis we propose an experimental method for performance evaluation of distributed systems based on the replay of production workload trace extracts. These extracts, replaced in their original context, enable to experiment the change of configuration of the system in an online workload and observe the different configurations results. Our technical contribution in this experimental approach is twofold. We propose a first tool to construct the environment in which the experimentation will take place, then we propose a second set of tools that automatize the experiment setup and that replay the trace extract within its original context.Finally, these contributions conducted together, enable to gain a better knowledge of HPC platforms. As future works, the approach proposed in this thesis will serve as a basis to further study larger infrastructures

High Performance Computing is preparing the era of the transition from Petascale to Exascale. Distributed computing systems are already facing new scalability problems due to the increasing number of computing resources to manage. It is now necessary to study in depth these systems and comprehend their behaviors, strengths and weaknesses to better build the next generation. The complexity of managing users applications on the resources conducted to the analysis of the workload the platform has to support, this to provide them an efficient service. The need for workload comprehension has led to the collection of traces from production systems and to the proposal of a standard workload format. These contributions enabled the study of numerous of these traces. This also led to the construction of several models, based on the statistical analysis of the different workloads from the collection. Until recently, existing workload traces did not enabled researchers to study the consumption of resources by the jobs in a temporal way. This is now changing with the need for characterization of jobs consumption patterns. In the first part of this thesis we propose a study of existing workload traces. Then we contribute with an observation of cluster workloads with the consideration of the jobs resource consumptions over time. This highlights specific and unattended patterns in the usage of resources from users. Finally, we propose an extension of the former standard workload format that enables to add such temporal consumptions without loosing the benefit of the existing works. Experimental approaches based on workload models have also served the goal of distributed systems evaluation. Existing models describe the average behavior of observed systems. However, although the study of average behaviors is essential for the understanding of distributed systems, the study of critical cases and particular scenarios is also necessary. This study would give a more complete view and under- standing of the performance of resource and job management. In the second part of this thesis we propose an experimental method for performance evaluation of distributed systems based on the replay of production workload trace extracts. These extracts, replaced in their original context, enable to experiment the change of configuration of the system in an online workload and observe the different configurations results. Our technical contribution in this experimental approach is twofold. We propose a first tool to construct the environment in which the experi- mentation will take place, then we propose a second set of tools that automatize the experiment setup and that replay the trace extract within its original context. Finally, these contributions conducted together, enable to gain a better knowledge of HPC platforms. As future works, the approach proposed in this thesis will serve as a basis to further study larger infrastructures.

Cette thèse se place dans le cadre de la vérification automatique des systèmes distribués. Elle aborde le problème de la spécification pour de tels systèmes, qui consiste à définir un formalisme logique pour décrire des propriétés des comportements de systèmes. On en attend qu'il soit facile d'exprimer les propriétés courantes (accessibilité, sûreté, exclusion mutuelle, vivacité, etc.). On souhaite par ailleurs que la vérification de ces propriétés soient aisée. Il s'agit donc de trouver un compromis entre pouvoir d'expression et simplicité d'utilisation.

On s'intéresse ensuite à la modélisation des systèmes concurrents, en recherchant à nouveau un compromis entre réalisme des modèles et facilité de vérification. Les modèles étudiés dans ce travail sont les automates asynchrones, qui modélisent des processus concurrents communiquant par mémoire partagée.

La thèse s'intéresse enfin au problème de la synthèse de contrôleur. Étant donné un système spécifié de façon incomplète, donc non-déterministe, en interaction avec un environnement, il s'agit de calculer de manière automatique comment restreindre son comportement afin qu'il vérifie une spécification donnée (quelles que soient les actions de l'environnement). Ce problème se formule en
termes de jeux. Dans le cas distribué, les jeux ont naturellement plusieurs joueurs. Dans ce cadre, la plupart des résultats sont négatifs : il est indécidable de savoir si on peut ou non contrôler un tel système. Cette thèse prouve que certaines propriétés de l'architecture de communication garantissent décidabilité pour toute spécification régulière.

Actuellement, les entreprises, les organisations et les fournisseurs de services ont besoin de publier leurs services et de les rendre accessibles à leurs clients. Les clients ont besoin de découvrir les services et de choisir ceux qui répondent à leurs exigences. Dans ce contexte, il s'agit de décrire les services le plus précisément possible afin qu'un client puisse trouver le service désiré. La description de services par des propriétés devient donc importante pour l'interrogation et la sélection de services et doit être prise en compte dans l'étape de l'indexation et de la publication. Les propriétés de services peuvent être utilisées par les fournisseurs (entreprises, organisations, etc. ) afin de publier leurs services. D'autre part, pour rechercher un service, les clients doivent disposer des approches leur permettent de découvrir les services en interrogeant leurs propriétés et en donnant la valeur désirée de chacune d'entre elles. Beaucoup d'approches ont été réalisées dans ce contexte, citons par exemple ODP trader et OMG CORBA trader, UDDI de Web services, etc. Dans cette thèse, nous proposons un modèle de méta données pour la description d'un service. Il peut être utilisé par les clients/serveurs pour publier/interroger un service. Il comporte trois niveaux de description : propriétés statiques, comportement et interface. Nous utilisons les automates pour décrire le comportement d'un service. Nous développons un trader en utilisant les ontologies. Il permet la recherche de services d'une manière flexible en utilisant la logique du premier ordre. Nous étendons ce trader pour adresser le comportement du service ainsi que son interface. À partir de cette description de service, sous plusieurs vues, nous pouvons récupérer des informations sur les propriétés, l'interface et le comportement sur les services enregistrés. Ceci nous permet alors de composer des services et d'en créer de nouveaux. Dans ce contexte, nous proposons une approche sur la composition de services qui permet de combiner et d'assembler plusieurs services pour satisfaire les demandes de clients. Le trader étendu comporte aussi une approche pour adapter les services selon les exigences de clients
Currently, the companies, the organizations and the services providers need to publish their services and to make them available to their clients. The clients need to discover the services and to select those that satisfy their requirements. In this context, services must be described as precisely as possible so that a client can find the service desired. The service description by properties becomes important to query and to select services and it must be taken into account in the indexing and publication stage. The providers (companies, organizations, etc. ) can use the services properties to publish their services. On the other side the clients, to discover a service, must dispose approaches that allow them to discover services by querying their properties and by assigning desired values for each property. In this context, many approaches have been developed like ODP trader and OMG CORBA, UDDI of services Web, etc. In this dissertation, we propose a meta data model for service description. It can be used by clients/servers to query/publish a service. It contains three levels of description : static properties, behavior and interface. We use automata to describe the service behavior. We design and implement a trader based on ontologies. It permits the service discovery in a flexible and expressive way by using the logic of the first order. We extend this trader to address the behavior of a service and its interface. Based on this description, under several views, we can get information about the properties, the interface and the behavior on the services stored. This allows us to compose services in order to create novel ones. In this context, we propose an approach for services composition that allows us to combine and collect service to satisfy the client's requests. The trader extended comprises also an approach for adapting service to the client's requirements

Ensuring the health of a distributed system with frequent updates is complicated. Many tools exist to improve developers’ comprehension and productivity in this task, but room for improvement exists. Based on previous research within request flow comparison, we propose a system design for using distributed tracing data in the process of reviewing code changes. The design is evaluated from the perspective of system performance and developer productivity using a critical production system at a large software company. The results show that the design has minimal negative performance implications while providing a useful service to the developers. They also show a positive but statistically insignificant effect on productivity during the evaluation period. To a large extent, developers adopted the tool into their workflow to explore and improve system understanding. This use case deviates from the design target of providing a method to compare changes between software versions. We conclude that the design is successful, but more optimization of functionality and a higher rate of adoption would likely improve the effects the tool could have.
Att säkerställa stabilitet i ett distribuerat system med hög frekvens av uppdateringar är komplicerat. I dagsläget finns många verktyg som hjälper utvecklare i deras förståelse och produktivitet relaterat till den här typen av problem, dock finns fortfarande möjliga förbättringar. Baserat på tidigare forskning inom teknik för att jämföra protokollförfrågningsflöden mellan mjukvaruversioner så föreslår vi en systemdesign för ett nytt verktyg. Designen använder sig av data från distribuerad tracing för att förbättra arbetsflödet relaterat till kodgranskning. Designen utvärderas både prestanda och produktivitetsmässigt under utvecklingen av ett affärskritiskt produktionssystem på ett stort mjukvaruföretag. Resultaten visar att designen har mycket låg inverkan på prestandan av systemet där det införs, samtidigt som den tillhandahåller ett användbart verktyg till utvecklarna. Resultaten visar också på en positiv men statistiskt insignifikant effekt på utvecklarnas produktivitet. Utvecklarna använde primärt verktyget för att utforska och förbättra sin egen förståelse av systemet som helhet. Detta användningsområde avvek från det ursprungliga målet med designen, vilket var att tillhandahålla en tjänst för att jämföra mjukvaruversioner med varandra. Från resultaten drar vi slutsatsen att designen som helhet var lyckad, men mer optimering av funktionalitet och mer effektivt införande av verktyget i arbetsflödet hade troligtvis resulterat i större positiva effekter på organisationen.

The process of forming a long-lasting memory may involve the selective linking together of neural representations stored widely throughout neocortex. The successful binding together of these disparate representations may require their coordinated reactivation while the cortex is 'offline' i.e., not engaged in processing external stimuli. This hypothesis was tested through simultaneous extracellular recording of 28-99 cells over four sites in the macaque neocortex. The recordings were conducted as the monkey performed repetitive reaching tasks, and in rest periods immediately preceding and following the task. In motor, somatosensory and parietal cortex (but not prefrontal cortex), the task-related neural activity patterns within and across regions were similar to the activity patterns seen afterwards, during the rest epoch. Moreover, the temporal sequences of neural ensemble activity that occurred during task performance were preserved in subsequent rest. The preservation of correlation structure and temporal sequencing are consistent with the reactivation of a memory trace and not merely the persistence of a fixed activity pattern. The observed memory trace reactivation was coordinated over large expanses of neocortex, confirming a fundamental tenet of the trace replay theory of memory consolidation.

L'apprentissage des matrices de rang faible est un problème de grande importance dans les statistiques, l'apprentissage automatique, la vision par ordinateur et les systèmes de recommandation. En raison de sa nature NP-difficile, une des approches principales consiste à résoudre sa relaxation convexe la plus étroite : la minimisation de la norme de trace. Parmi les différents algorithmes capables de résoudre cette optimisation, on peut citer la méthode de Frank-Wolfe, particulièrement adaptée aux matrices de grande dimension. En préparation à l'utilisation d'infrastructures distribuées pour accélérer le calcul, cette étude vise à explorer la possibilité d'exécuter l'algorithme de Frank-Wolfe dans un réseau en étoile avec le modèle BSP (Bulk Synchronous Parallel) et à étudier son efficacité théorique et empirique. Concernant l'aspect théorique, cette étude revisite le taux de convergence déterministe de Frank-Wolfe et l'étend à des cas non déterministes. En particulier, il montre qu'avec le sous-problème linéaire résolu de manière appropriée, Frank-Wolfe peut atteindre un taux de convergence sous-linéaire à la fois en espérance et avec une probabilité élevée. Cette contribution pose la fondation théorique de l'utilisation de la méthode de la puissance itérée ou de l'algorithme de Lanczos pour résoudre le sous-problème linéaire de Frank-Wolfe associé à la minimisation de la norme de trace. Concernant l'aspect algorithmique, dans le cadre de BSP, cette étude propose et analyse quatre stratégies pour le sous-problème linéaire ainsi que des méthodes pour la recherche linéaire. En outre, remarquant la propriété de mise à jour de rang-1 de Frank-Wolfe, il met à jour le gradient de manière récursive, avec une représentation dense ou de rang faible, au lieu de le recalculer de manière répétée à partir de zéro. Toutes ces conceptions sont génériques et s'appliquent à toutes les infrastructures distribuées compatibles avec le modèle BSP. Concernant l'aspect empirique, cette étude teste les conceptions algorithmiques proposées dans un cluster Apache SPARK. Selon les résultats des expériences, pour le sous-problème linéaire, la centralisation des gradients ou la moyenne des vecteurs singuliers est suffisante dans le cas de faible dimension, alors que la méthode de la puissance itérée distribuée, avec aussi peu qu'une ou deux itérations par époque, excelle dans le cas de grande dimension. La librairie Python développée pour les expériences est modulaire, extensible et prête à être déployée dans un contexte industriel. Cette étude a rempli sa fonction de preuve de concept. Suivant le chemin qu'il met en place, des solveurs peuvent être implémentés pour différentes infrastructures, parmi lesquelles des clusters GPU, pour résoudre des problèmes pratiques dans des contextes spécifiques. En outre, ses excellentes performances dans le jeu de données ImageNet le rendent prometteur pour l'apprentissage en profondeur
Learning low-rank matrices is a problem of great importance in statistics, machine learning, computer vision, recommender systems, etc. Because of its NP-hard nature, a principled approach is to solve its tightest convex relaxation : trace norm minimization. Among various algorithms capable of solving this optimization is the Frank-Wolfe method, which is particularly suitable for high-dimensional matrices. In preparation for the usage of distributed infrastructures to further accelerate the computation, this study aims at exploring the possibility of executing the Frank-Wolfe algorithm in a star network with the Bulk Synchronous Parallel (BSP) model and investigating its efficiency both theoretically and empirically. In the theoretical aspect, this study revisits Frank-Wolfe's fundamental deterministic sublinear convergence rate and extends it to nondeterministic cases. In particular, it shows that with the linear subproblem appropriately solved, Frank-Wolfe can achieve a sublinear convergence rate both in expectation and with high probability. This contribution lays the theoretical foundation of using power iteration or Lanczos iteration to solve the linear subproblem for trace norm minimization. In the algorithmic aspect, within the BSP model, this study proposes and analyzes four strategies for the linear subproblem as well as methods for the line search. Moreover, noticing Frank-Wolfe's rank-1 update property, it updates the gradient recursively, with either a dense or a low-rank representation, instead of repeatedly recalculating it from scratch. All of these designs are generic and apply to any distributed infrastructures compatible with the BSP model. In the empirical aspect, this study tests the proposed algorithmic designs in an Apache SPARK cluster. According to the experiment results, for the linear subproblem, centralizing the gradient or averaging the singular vectors is sufficient in the low-dimensional case, whereas distributed power iteration, with as few as one or two iterations per epoch, excels in the high-dimensional case. The Python package developed for the experiments is modular, extensible and ready to deploy in an industrial context. This study has achieved its function as proof of concept. Following the path it sets up, solvers can be implemented for various infrastructures, among which GPU clusters, to solve practical problems in specific contexts. Besides, its excellent performance in the ImageNet dataset makes it promising for deep learning

Many components of distributed systems are now designed so that they emit alarms when some particular conditions are met. Nevertheless inferring the causes of the failures remains a challenging problem which sometimes requires to reconstruct a part of the history of the system. We propose to compute the explanations of an observation, according to a model of the net. For this we use the notion of unfoldings of Petri nets. Unfoldings provide an efficient way to represent the executions and to emphasize the causality and concurrency relations between the events. This allows to find easily the causes of the failures. But in order to model real systems, high-level extensions of Petri nets are often required. We define symbolic unfoldings of colored Petri nets, dynamic nets and time Petri nets. In symbolic unfoldings the histories are grouped into families of parameterized executions
De nombreux composants des systèmes répartis sont maintenant conçus pour émettre des alarmes lorsque certaines situations prédéfinies sont détectées. Malgré cela, inférer les causes des pannes à partir des alarmes reste un problème difficile, qui nécessite parfois de reconstituer une partie de l'histoire du système. Nous calculons les explications d'une observation en nous appuyant sur un modèle du système et sur la notion de dépliages de réseaux de Petri. Les dépliages fournissent une représentation efficace des exécutions en mettant en valeur les relations de causalité et de concurrence entre les événements. Ceci permet de remonter facilement aux causes des pannes. Mais pour modéliser de vrais systèmes, il est souvent indispensable d'utiliser des extensions de haut niveau des réseaux de Petri. Nous définissons des dépliages symboliques pour les réseaux de Petri colorés, les réseaux dynamiques et les réseaux de Petri temporels, en identifiant des familles d'exécutions paramétrées

La conception des systèmes multimédia embarqués présente de nombreux défis comme la croissante complexité du logiciel et du matériel sous-jacent, ou les pressions liées aux délais de mise en marche. L'optimisation du processus de débogage et validation du logiciel peut aider à réduire sensiblement le temps de développement. Parmi les outils de débogage de systèmes embarqués, un puissant outil largement utilisé est l'analyse de traces d'exécution. Cependant, l'évolution des techniques de tra¸cage dans les systèmes embarqués se traduit par des traces d'exécution avec une grande quantité d'information, à tel point que leur analyse manuelle devient ingérable. Dans ce cas, les techniques de recherche de motifs peuvent aider en trouvant des motifs intéressants dans de grandes quantités d'information. Concrètement, dans cette thèse, nous nous intéressons à la découverte de comportements périodiques sur des applications multimédia. Donc, les contributions de cette thèse concernent l'analyse des traces d'exécution d'applications multimédia en utilisant des techniques de recherche de motifs périodiques fréquents. Concernant la recherche de motifs périodiques, nous proposons une définition de motif périodique adaptée aux caractéristiques de la programmation paralléle. Nous proposons ensuite une représentation condensée de l'ensemble de motifs périodiques fréquents, appelée Core Periodic Concepts (CPC), en adoptant une approche basée sur les relations triadiques. De plus, nous définissons quelques propriétés de connexion entre ces motifs, ce qui nous permet de mettre en oeuvre un algorithme efficace de recherche de CPC, appelé PerMiner. Pour montrer l'efficacité et le passage à l'échelle de PerMiner, nous réalisons une analyse rigoureuse qui montre que PerMiner est au moins deux ordres de grandeur plus rapide que l'état de l'art. En plus, nous réalisons un analyse de l'efficacité de PerMiner sur une trace d'exécution d'une application multimédia réelle en présentant l'accélération accompli par la version parallèle de l'algorithme. Concernant les systèmes embarqués, nous proposons un premier pas vers une méthodologie qui explique comment utiliser notre approche dans l'analyse de traces d'exécution d'applications multimédia. Avant d'appliquer la recherche de motifs fréquents, les traces d'exécution doivent ˆetre traitées, et pour cela nous proposons plusieurs techniques de pré-traitement des traces. En plus, pour le post-traitement des motifs périodiques, nous proposons deux outils : un outil qui trouve des pairs de motifs en compétition ; et un outil de visualisation de CPC, appelé CPCViewer. Finalement, nous montrons que notre approche peut aider dans le débogage des applications multimédia à travers deux études de cas sur des traces d'exécution d'applications multimédia réelles.

Since the advent of the computer in the late 1950s, scientists and engineers have pushed the limits of the computing power available to them to solve physical problems via computational simulations. Early computer languages evaluated program logic in a sequential manner, thereby forcing the designer to think of the problem solution in terms of a sequential process. Object-oriented analysis and design have introduced new concepts for solving systems of engineering problems. The term object-oriented was first introduced by Alan Kay [1] in the late 1960s; however, mainstream incorporation of object-oriented programming did not occur until the mid- to late 1990s. The principles and methods underlying object-oriented programming center around objects that communicate with one another and work together to model the physical system. Program functions and data are grouped together to represent the objects. This dissertation extends object-oriented modeling concepts to model algorithms in a generic manner for solving interconnected, multi-domain problems. This work is based on an extension of Graph Trace Analysis (GTA) which was originally developed in the 1990â s for power distribution system design. Because of GTAâ s ability to combine and restructure analysis methodologies from a variety of problem domains, it is now being used for integrated power distribution and transmission system design, operations and control. Over the last few years research has begun to formalize GTA into a multidiscipline approach that uses generic algorithms and a common model-based analysis framework. This dissertation provides an overview of the concepts used in GTA, and then discusses the main problems and potential generic algorithm based solutions associated with design and control of interdependent reconfigurable systems. These include: â ¢ Decoupling analysis into distinct component and system level equations. â ¢ Using iterator based topology management and algorithms instead of matrices. â ¢ Using composition to implement polymorphism and simplify data management. â ¢ Using dependency components to structure analysis across different systems types. â ¢ Defining component level equations for power, gas and fluid systems in terms of across and though variables. This dissertation presents a methodology for solving interdependent, multi-domain networks with generic algorithms. The methodology enables modeling of very large systems and the solution of the systems can be accomplished without the need for matrix solvers. The solution technique incorporates a binary search algorithm for accelerating the solution of looped systems. Introduction of generic algorithms enables the system solver to be written such that it is independent of the system type. Example fluid and electrical systems are solved to illustrate the generic nature of the approach.
Ph. D.

This thesis aims to improve the cross-platform reliability of software components tested on distributed CI systems. More specifically, it is centered on extracting crash information from cross-platform crashes. Crash information was generated and parsed on Mac, Linux and Windows. The crash information proved to be valuable for developers in their day-to-day job, especially the raw crash information. However, the graph visualizations that were created out of this information proved to be less than satisfactory for developers.

The main objective of this thesis is to characterize hydrological processes in a Mediterranean mountain catchment, by combining distributed hydrological measurements and environmental tracers in order to improve the understanding of catchment hydrological function. During the period 2009-2013 hydrological data were collected from the Vallcebre Research Catchments, monitored since 1996 by the Surface Hydrology and Erosion group of the IDAEA-CSIC. Data include, in addition to rainfall and discharge measurements, distributed hydrological measurements and environmental tracers (both geochemical and isotope ones) at different time scales (seasonal to event scale). With this information, this study first investigates the spatial and temporal variability of the depth to water table during rainfall-runoff events. The results show that the depth to water table did not rise uniformly throughout the catchment during rainfall-runoff events. The spatial variability of depth to water table was mainly controlled by location characteristics, especially the piezometer distance from the stream, which influenced the distribution of wetness conditions within the catchment. The wetness conditions in turn affected the timing of the water table response, as well as the magnitude of the streamflow response. Spatio-temporal water table variability during floods varied, depending on the catchment's antecedent wetness conditions Dissolved organic carbon (DOC) concentration dynamics in different hydrological compartments were also analysed (rainfall, soil water, groundwater and stream water) at different time scales (seasonal to event scale). The results show some seasonality in rainwater and soil water DOC concentrations, while no clear seasonality was found in stream water and groundwater, where DOC dynamics were strongly related to discharge and water table variations. During storm events, DOC concentrations increased systematically in stream water. In addition, for storm events with several discharge peaks, the slope of the discharge/DOC concentration relationship was higher for the first peak. The increase in stream water DOC concentration during floods suggested a relevant contribution of soil water, as well as the existence of stream water DOC sources near or in the stream bed. The rather similar dynamics of stream water DOC concentration in all floods contrasted with the diversity of hydrological processes observed. This raises the question of the origin of the rapid DOC increase found and the validity of the use of DOC as a tracer. Finally, water mean transit time (MTT) was calculated in different hydrological compartments of the catchment, using stable isotopes (δ18O and δ2H) and tritium. The use of δ18O signal variations in rainfall and in the sampled hydrological compartments showed some limitations on water age calculation in the catchment studied: it only indicated that MTT was greater than two years. The use of a new methodology (TEPMGLUE) to calculate MTT using tritium allowed consideration of different sources of uncertainty in water age determination, as well as evaluation of the benefit of using samples of different ages and of differing analytical quality. The results showed similar MTT calculations, whether including only the water samples taken in the 1990s or using all samples (1996-1998 and 2013). However, when calculating MTT with only high analytical quality samples taken in 2013, two different MTTs were obtained. The MTT results showed that, in the Vallcebre catchments, well water was the youngest, followed by stream and spring water. The study also showed the relevance of the rainfall tritium input function to MTT calculations. Finally, results showed that topography did not affect MTT spatial distribution, whereas geological settings did.
L'objectiu de la tesi doctoral és millorar la caracterització dels processos hidrològics de les conques mediterrànies de muntanya utilitzant mesures hidrològiques distribuïdes i traçadors ambientals, amb la finalitat de comprendre'n el funcionament hidrològic. Durant el període 2009-2013 es van recollir dades hidrològiques de les conques d'investigació de Vallcebre, monitoritzades des de 1996 pel grup d’Hidrologia Superficial i Erosió del IDAEA-CSIC. Les dades inclouen, a més a més de registres de precipitacions i cabal, mesures hidrològiques distribuïdes i de dades de traçadors ambientals (geoquímics i isotòpics) a diferents escales temporals (estacional i d'esdeveniment). A partir d’aquesta informació s'ha analitzat, en primer lloc, la dinàmica espaciotemporal del nivell freàtic durant els episodis de pluja. Els resultats mostren que, durant aquests episodis, el nivell freàtic no reacciona de manera homogènia a tota la conca. La variabilitat espaial del nivell freàtic està molt afectada per les característiques de la localització de cada piezòmetre, especialment la distància al torrent, que té un paper essencial en la distribució de les condicions d'humitat de la conca. Aquestes condicions d'humitat, a la vegada, afecten al temps de resposta del nivell freàtic i a la magnitud de la resposta hidrològica. Segons les condicions d'humitat de la conca, la variabilitat espaciotemporal del nivell freàtic durant esdeveniments és diferent. En condicions seques, la variabilitat espaciotemporal del nivell freàtic és alta al llarg de tot l'esdeveniment. L'escolament es genera principalment per la contribució de les àrees que estan a prop del torrent. En canvi, en condicions de transició i especialment en condicions d'humitat, la variabilitat espaciotemporal del nivell freàtic és menor i respon de forma més ràpida, i produeix una resposta hidrològica major a escala de conca. Tot seguit, s'ha analitzat la dinàmica de la concentració de carboni orgànic dissolt (DOC) en diferents compartiments hidrològics (aigua de pluja, aigua del sòl, aigua subterrània i aigua del torrent) i a diferents escales temporals (estacional i d'esdeveniment). Els resultats mostren una variació estacional de la concentració de DOC a l'aigua de pluja i a la del sòl; però no s'aprecia aquesta estacionalitat a l'aigua subterrània i a la del torrent, a causa de que la concentració de DOC als dos compartiments està relacionada amb la dinàmica del nivell freàtic i la del cabal. A escala d’esdeveniment s'observa un augment sistemàtic de la concentració de DOC al torrent. A més a més, en els esdeveniments de diversos pics de cabal el pendent de la relació cabal/concentració de DOC sempre és major al primer pic. Aquest augment de la concentració de DOC al cabal durant els esdeveniments suggereix una contribució rellevant de l’aigua del sòl, però també l'existència de fonts de DOC pròximes al torrent o del mateix llit del torrent. D'altra banda, la relativa semblança de la dinàmica del DOC en el torrent durant esdeveniments de diferent magnitud i amb processos hidrològics contrastats molt diferents, planteja la qüestió de l'origen de l'augment de DOC i el seu possible ús com a traçador. Per últim, s'ha realitzat el càlcul del temps mitjà de trànsit (MTT) de l'aigua als diferents compartiments hidrològics de la conca utilitzant, primer, els isòtops estables (δ18O i δ2H), i segon, el triti. L'ús de la variació del senyal isotòpic del δ18O a l'aigua de pluja i a diferents compartiments d'aigua mostrejats presenta limitacions a la conca d'estudi; únicament permet demostrar que el MTT de les aigües és major de dos anys. L'ús d'una nova metodologia (TEPMGLUE) per calcular el MTT, utilitzant el triti, permet considerar diferents fonts d'incertesa i avaluar l'ús de mostres d'edats diferents i de diferent precisió analítica. Els resultats revelen que s'obtenen valors de MTT similars, tant utilitzant mostres dels anys noranta, com totes les mostres disponibles (1996-1998 i 2013). En canvi, quan únicament es calcula el MTT utilitzant les mostres del 2013 s'obtenen dos valors de MTT molt diferents. Els resultats de l’estudi també mostren que a les conques de Vallcebre l'aigua més jove és la dels pous, seguida de l'aigua del torrents i la de les fonts. L'estudi també demostra el paper rellevant de la funció d'entrada de triti amb la pluja per calcular el MTT. Per últim, els resultats suggereixen que la topografia no afecta la distribució espaial del MTT de la conca, però sí la geologia. Tots els resultats d'aquesta tesi aporten informació addicional sobre el funcionament hidrològic de les conques de Vallcebre, i a més a més suggereixen noves línies d'investigació
El objetivo de esta tesis doctoral es mejorar la caracterización de los procesos hidrológicos de las cuencas Mediterráneas de montaña, utilizando mediciones hidrológicas distribuidas y trazadores ambientales, con la finalidad de comprender mejor su funcionamiento hidrológico. Durante el período 2009-2013 se recogieron datos hidrológicos de las cuencas de investigación de Vallcebre, monitorizadas desde 1996 por el grupo de Hidrología superficial y Erosión del IDAEA-CSIC. Los datos incluyen, además del registro de precipitaciones y caudales, mediciones hidrológicas distribuidas y datos de trazadores ambientales (geoquímicos e isotópicos) a diferentes escalas temporales (estacional y de crecida). Con esta información, se ha analizado en primer lugar la dinámica espacio-temporal del nivel freático durante las crecidas. Los resultados muestran que durante las crecidas el nivel freático no reacciona de forma homogénea en toda la cuenca. La variabilidad espacial del nivel freático está muy afectada por las características de la localización de cada piezómetro, especialmente la distancia al cauce que tiene un papel esencial en la distribución de las condiciones de humedad de la cuenca. Estas condiciones de humedad, a su vez, afectan al tiempo de respuesta del nivel freático y a la magnitud de la respuesta hidrológica. Según las condiciones de humedad de la cuenca la variabilidad espacio-temporal del nivel freático durante la crecida es diferente. En condiciones secas la variabilidad espacio-temporal del nivel freático es alta a lo largo de toda la crecida. En estas condiciones, la escorrentía se genera principalmente por la contribución de las áreas próximas al cauce. En cambio, en condiciones de transición y especialmente en condiciones húmedas, la variabilidad espacio-temporal del nivel freático es menor y su respuesta es mas rápida, produciendo una mayor respuesta hidrológica en la cuenca. Seguidamente, se ha analizado la dinámica de la concentración del carbono orgánico disuelto (DOC) en diferentes compartimentos (agua de lluvia, agua del suelo, agua subterránea y agua del cauce) a diferentes escalas temporales (estacional y de crecida). Los resultados indican una variación estacional de la concentración de DOC en el agua de lluvia y en la del suelo. Sin embargo, no se aprecia esta estacionalidad en el agua subterránea y en la del cauce, debido a que la concentración de DOC en estos compartimentos está relacionada con la dinámica del nivel freático y del caudal. Durante las crecidas se observa un aumento sistemático de la concentración de DOC en el cauce. Además en las crecidas con varios picos de caudal el pendiente de la relación caudal/concentración de DOC es siempre mayor en el primer pico. El aumento de la concentración de DOC en el cauce durante las crecidas sugiere una contribución relevante del agua del suelo, pero también la existencia de fuentes de DOC próximas al cauce o en el mismo lecho. La relativa similitud de la dinámica del DOC en el cauce durante eventos de distinta magnitud y con procesos hidrológicos contrastados muy distintos, plantea la cuestión del origen del aumento de DOC y su posible utilización como trazador. Por último, se ha realizado el cálculo del tiempo medio de transito (MTT) del agua, en diferentes compartimentos hidrológicos de la cuenca utilizando, primero, isótopos estables (δ18O y δ2H), y después el tritio. El uso de la variación de la señal isotópica del δ18O en el agua de lluvia y en los diferentes compartimentos de agua muestreados muestra sus limitaciones en la cuenca de estudio; solo permite demostrar que el MTT es mayor de dos años. El uso de una nueva metodología (TEPMGLUE) para calcular el MTT usando el tritio, permite considerar diferentes fuentes de incertidumbre y evaluar el uso de muestras de edades diferentes y con distinta precisión analítica. Los resultados revelan que se obtienen valores de MTT similares, tanto utilizando muestras de los años noventa, como todas las muestras disponibles (1996-1998 y 2013). Sin embargo, cuando se calcula el MTT utilizando solo muestras del 2013 se obtienen dos valores de MTT muy distintos. Los resultados de MTT muestran que en las cuencas de Vallcebre el agua es más jóven en los pozos, seguida del agua de los cauces y la de las fuentes. El estudio también demuestra el papel relevante de la función de entrada de tritio en la lluvia para el cálculo del MTT. Por último, los resultados sugieren que la topografía no afecta la distribución espacial del MTT en la cuenca, pero sí la geología. Todos los resultados de esta tesis aportan información adicional sobre el funcionamiento hidrológico de las cuencas de Vallcebre, además sugieren nuevas líneas de investigación.

This work presents an alternative approach to power system computations, Graph Trace Analysis (GTA), and applies GTA to the power flow problem. A novel power flow algorithm is presented, where GTA traces are used to implement a modified Gauss-Seidel algorithm coupled with a continuation method. GTA is derived from the Generic Programming Paradigm of computer science. It uses topology iterators to move through components in a model and perform calculations. Two advantages that GTA brings are the separation of system equations from component equations and the ability to distribute calculations across processors. The implementation of KVL and KCL in GTA is described. The GTA based power flow algorithm is shown to solve IEEE standard transmission models, IEEE standard distribution models, and integrated transmission and distribution models (hybrid models) constructed from modifying IEEE standard models. The GTA power flow is shown to solve a set of robustness testing circuits, and solutions are compared with other power flow algorithms. This comparison illustrates convergence characteristics of different power flow algorithms in the presence of voltage stability concerns. It is also demonstrated that the GTA power flow solves integrated transmission and distribution system models. Advantages that GTA power flow bring are the ability to solve realistic, complex circuit models that pose problems to many traditional algorithms; the ability to solve circuits that are operating far from nominal conditions; and the ability to solve transmission and distribution networks together in the same model.
PHD

Atualmente, muitas redes com características dinâmicas estão em funcionamento (por exemplo MANETs, DTNs, redes oportunistas, etc). Neste trabalho, estudamos um modelo para estas redes chamado de Grafos Evolutivos, que permite expressar a dinamicidade das conexões entre nós por meio de uma simples extensão da estrutura comum de grafos. Esta modelagem é utilizada no arcabouço proposto por Casteigts et al. para definir algoritmos distribuídos em redes dinâmicas, que utiliza grafos evolutivos para representar a topologia da rede e renomeação de rótulos para expressar a comunicação entre os nós. Utilizamos esta abordagem para estudar o problema da exclusão mútua distribuída em redes dinâmicas e diversos algoritmos propostos para ele, a fim de definir e validar suas condições necessárias e suficientes de conectividade em redes dinâmicas. Além da formalização de algoritmos, o modelo de grafos evolutivos também pode ser utilizado para analisar redes dinâmicas. Rastros de redes dinâmicas reais são amplamente utilizados na literatura para estudos de algoritmos pois estes geram resultados mais realísticos do que redes simuladas com padrões de movimento. A partir dos detalhes de cada conexão entre nós de um destes rastros, é possível construir um grafo evolutivo, do qual se pode extrair dados como jornadas ótimas entre nós, variação da conectividade no tempo, estabilidade, e periodicidade. Com as informações mencionadas, um pesquisador pode observar com maior precisão as características do rastro, o que facilita na escolha da rede mais apropriada para sua necessidade. Além disso, o conhecimento prévio de tais características de uma rede auxilia no estudo do comportamento de algoritmos executados sobre ela e provém uma validação para suposições geralmente feitas pelos pesquisadores. Para fornecer estas informações, desenvolvemos uma ferramenta Web que analisa rastros de redes dinâmicas e agrega os dados em um formato de fácil visualização. Descrevemos, neste trabalho, a implementação e a utilidade de todos os serviços da ferramenta.
Lately, several networks with dynamic properties (for instance MANETs, DTNs, opportunistic networks, etc) are functioning. In this work, we studied a model for these networks called Evolving Graphs, which allows the expression of the dynamicity of the conections between nodes through a simple extension of the common graph structure. This model is used by the framework proposed by Casteigts et al. to define distributed algorithms in dynamic networks, which uses evolving graphs to represent the network topology and graph relabelling to express the communication between nodes. Using this approach, we study the distributed mutual exclusion problem in dynamic networks and several algorithms proposed to solve it, in order to define and validate their necessary and sufficient connectivity conditions. Apart from the formalization of algorithms, the evolving graphs model can also be used to analyze dynamic networks. Dynamic network traces are widely used in the literature in order to study algorithms, as they generate better results than simulated networks with movement patterns. From the details of every connection between nodes in a trace, it is possible to build an evolving graph, from which a large amount of information can be extracted, such as optimal journeys between nodes, variation of the conectivity over time, stability and periodicity. With the aforementioned information, a researcher might observe the characteristics of a trace more precisely, which facilitates the process of choosing the most appropriate trace for his needs. Furthermore, the early knowledge of such characteristics of a network helps in the study of the behavior of the algorithms exected over it and provides a validation for the assumptions usually made by the researchers. In order to provide this information, we developed a web tool which analyzes dynamic network traces and aggregates the data in an easily readable format. In this work, we describe the implementation and usefulness of every service in the tool.

Lorsque l'on s’intéresse à l'activité humaine selon l'approche interactionniste de la cognition, la question de 'appropriation des dispositifs techniques par les utilisateurs est centrale. Selon cette posture, la cognition humaine est située dans la matérialité et distribuée entre les actants de la situation. En situation d'apprentissage, la question de 'appropriation des artefacts est majeure et renvoie en particulier aux situations où les apprenants mobilisent un environnement informatique pour l'apprentissage humain. Cependant, peu de recherches étudient les éléments de la situation favorisant l'appropriation du dispositif numérique et participer au développement de l’activité et du sujet agissant. Dans cette recherche, nous faisons l'hypothèse que la visualisation des traces informatiques des interactions entre sujet et système peut réduire l'écart qui existe entre l'utilisation du dispositif "prescrite" par le concepteur et l’utilisation "réelle" par l'utilisateur. Cet écart, gênant la genèse instrumentale de l'artefact, est souvent à l'origine de problèmes d'appropriation du dispositif. Cette recherche interroge le rôle de l'expérience en activité médiée et conjointe, et mobilise quatre domaines de recherche : l'interaction homme-machine, la psychologie de la cognition, l'activité médiée et la théorie de l'apprentissage. Nous proposons une mise en situation naturelle de co-rédaction instrumentalisée via un dispositif numérique composé de différents espaces communicationnels et rédactionnels. Nous réalisons une étude ethnométhodologique des interactions ayant lieu, entre sujets et entre sujets et espaces numériques, et du rôle des traces dans les interactions
Considering human activity according an interactionnist approach of cognition, a main question is the artefacts'appropriation by users. According to this approach, human cognition is situated in the materiality of the environment and distributed between people acting. In a learning situation, the question of appropriation is very important and concerns computer-mediated situations for learners. However, a few research works deal with situation elements supporting the numerical devices' appropriation, participating to the development of the activity and to the development of the learner. In this work, we hypothesize that visualisation of numerical traces of interactions between users and between users and the system can reduce the gap between the "prescribed" use of the device (by the designer) and the "real" use by users. This gap leads to appropriation problems, interfering with the instrumental genesis of the artefact. This work focuses on and interrogates the role of experience in a joint and mediated activity, and concerns four research domains : human-computer interaction, cognition psychology, mediated activity and learning theory. We study a natural co-writing activity, mediated by a numerical artefact composed of different communicationnal and editorial areas. With an ethnomethodological approach, we study interactions occuring between subjects and between subjets and numerical areas, and the role of traces in interactions

A new approach to the evaluation of the reliability of electrical systems is presented. In this approach a Graph Trace Analysis based approach is applied to integrated system models and reliability analysis. The analysis zones are extended from the traditional power system functional zones. The systems are modeled using containers with iterators, where the iterators manage graph edges and are used to process through the topology of the graph. The analysis provides a means of computationally handling dependent outages and cascading failures. The effects of adverse weather, time-varying loads, equipment age, installation environment, operation conditions are considered. Sequential Monte Carlo simulation is used to evaluate the reliability changes for different system configurations, including distributed generation and transmission lines. Historical weather records and loading are used to update the component failure rates on-the-fly. Simulation results are compared against historical reliability field measurements. Given a large and complex plant to operate, a real-time understanding of the networks and their situational reliability is important to operational decision support. This dissertation also introduces using an Integrated System Model in helping operators to minimize real-time problems. A real-time simulation architecture is described, which predicts where problems may occur, how serious they may be, and what is the possible root cause.
Ph. D.

Investigations into the neural correlates of memory have found the hippocampus to be a crucial structure for long-term declarative memories, but the exact nature of this contribution remains under debate. This paper covers three theories concerned with how the hippocampus is involved in long-term memory, namely the Standard Consolidation Model, the Multiple-Trace Theory, and the Distributed Reinstatement Theory. According to the Standard Consolidation Model, long-term declarative memories (both episodic and semantic) are dependent on the hippocampus for a limited time during which the memories undergo a process of consolidation, after which they become dependent on the neocortex. In contrast, the Multiple-Trace Theory argues that detailed and context-specific episodic (but not semantic) memories remain dependent on the hippocampus indefinitely. While both the aforementioned theories posit that memories are initially dependent on the hippocampus, the Distributed Reinstatement Theory does not. Advocates of this theory propose that several memory systems compete for the encoding of a memory, and that the hippocampus usually is the dominant system. However, it is also suggested that the other (unspecified) memory systems can overcome the hippocampal dominance through extensive and distributed learning sessions. In this paper, findings from both human and rodent studies focusing on the hippocampus are reviewed and used to evaluate the claims made by each theory on a systems level.

Pour la gestion de grands systèmes distribués (réseaux de
télécommunications par exemple) il est utile d'étudier des
modèles de concurrence sous la sémantique de traces. Dans
cette optique, on propose une extension probabiliste des
structures d'événements et des réseaux de Petri 1-bornés
(réseaux markoviens).
On prouve un propriété de Markov forte pour ces modèles,
et on donne des applications à la récurrence des réseaux.
On montre une Loi forte des grands nombres pour
les réseaux récurrents et suffisement synchrones,
avec applications a l'estimation statistique de
paramètres locaux.

Solar PV-based distributed generation has increased significantly over the last few years, and the rapid growth is expected to continue in the foreseeable future. As the penetration levels of distributed generation increase, power systems will become increasingly decentralized with bi-directional flow of electricity between the transmission and distribution networks. To manage such decentralized power systems, planners and operators need models that accurately reflect the structure of, and interactions between the transmission and distribution networks. Moreover, algorithms that can simulate the steady state and dynamics of power systems using these models are also needed. In this context, integrated transmission and distribution system modeling and simulation has become an important research area in recent years, and the primary focus so far has been on studying the steady state response of power systems using integrated transmission and distribution system models. The primary objective of this dissertation is to develop an analysis approach and a program that can simulate the dynamics of three phase, integrated transmission and distribution system models, and use the program to demonstrate the advantages of evaluating the impact of solar PV-based distributed generation on power systems dynamics using such models. To realize this objective, a new dynamic simulation analysis approach is presented, the implementation of the approach in a program is discussed, and verification studies are presented to demonstrate the accuracy of the program. A new dynamic model for small solar PV-based distributed generation is also investigated. This model can interface with unbalanced networks and change its real power output according to the incident solar irradiation. Finally, application of the dynamic simulation program for evaluating the impact of solar PV units using an integrated transmission and distribution system model is discussed. The dissertation presents a new approach for studying the impact of solar PV-based distributed generation on power systems dynamics, and demonstrates that the solar PV impact studies performed using the program and integrated transmission and distribution system models provide insights about the dynamic response of power systems that cannot be obtained using traditional dynamic simulation approaches that rely on transmission only models.
Ph. D.

The yield from a solar photovoltaic (PV) source is dependent on factors such as light and temperature. A control system called a maximum power point tracker (MPPT) ensures that the yield from a solar PV source is maximized in spite of these factors. This thesis presents a novel implementation of a perturb and observe (PO) MPPT. The implementation uses a switched capacitor step down converter and a custom digital circuit implementation of the PO algorithm. Working in tandem, the switched capacitor step down converter and the custom digital circuit implementation were able to successfully track the maximum power point of a simulated solar PV source. This implementation is free of the overhead encountered with general purpose processor based MPPT implementations. This makes this MPPT system a valid candidate for applications where general purpose processors are undesirable. This document will begin by discussing the current state of MPPT research. Afterward, this thesis will present studies done to be able to use the chosen switched capacitor step down converter. Then the digital circuit PO implementation will be discussed in detail. Simulations of the architecture will be presented. Finally, experimental validation using a hardware prototype will be shown.

L'objectif principal de ces travaux de thèse consiste à améliorer la compréhension de la structure interne des aquifères karstiques en s'appuyant sur l'interprétation de traçages artificiels. Pour répondre à cet objectif, cette thèse s'appuie sur les bassins versants karstiques du Baget et d'Aliou (Ariège, France) ainsi que d'autres hydrosystèmes karstiques (les systèmes de Norville, du Val de Loire et de Foubanne) inclus dans le Service National d'Observation des Karsts (SNO Karst, INSU-CNRS) et sur la base de données BD traçage du BRGM incluant notamment le système karstique de l'Ouysse. L'approche globale retenue pour ces travaux de thèse consiste à coupler les résultats de différentes approches. [1] La modélisation conceptuelle à réservoirs avec l'outil dédié KarstMod permet de partitionner les écoulements à l'échelle du bassin versant et d'appréhender la dynamique des échanges entre la matrice et les conduits, à court et long terme. On estime que la contribution annuelle de la matrice au débit à l'exutoire est d'environ 3% et qu'elle peut augmenter jusqu'à 25% en période de faible précipitation. [2] Une approche systémique de type "fonction de transfert" a été développée et appliquée à divers hydrosystèmes (dont les systèmes de Norville, de l'Ouysse, du Val de Loire et de Foubanne). Cette approche consiste à étudier la dynamique du transport de masse en soluté en interprétant des données de traçages artificiels et permet de partitionner les écoulements en estimant les temps de résidence des différentes composantes qui contribuent en transfert de solutés entre les points d'injection et de restitution. [3] Une approche distribuée intégrant un couplage de données géophysiques et de simulations pseudo-génétique permet de simuler des réseaux de drainage. Les écoulements et le transport de masse en soluté associé sont ensuite simulés et confrontés à des données de terrain. Cela constitue une approche originale pour déduire la structuration des écoulements souterrains à partir des données de traçages artificiels
The main objective of these thesis is to improve the understanding of the internal structure of karstic aquifers by relying on the interpretation of artificial tracer tests. To meet these objectives, this thesis is based on the karst watersheds of Baget and Aliou (Ariège, France) as well as other karst hydrosystems (the Norville, Val de Loire and Foubanne systems) included in the National Karts Observation Service (SNO Karst, INSU-CNRS) and the tracing database of the BRGM including the Ouysse karst system. The overall approach adopted for this thesis is to couple the results of different approaches. [1] Conceptual reservoirs modeling with the dedicated KarstMod model allows the partitioning of flows at the scale of the watershed and to understand the dynamics of the exchanges between the matrix and the conduits, in the short and long term. It is estimated that the annual contribution of the matrix to the total outflow is about 3% and can increase up to 25% during periods of low precipitation. [2] A systemic approach of the "transfer function" type has been developed and applied to various hydrosystems (the Norville, Ouysse, Val de Loire and Foubanne systems). This approach consists of studying the dynamics of solute mass transport by interpreting artificial tracer tests data and partitioning the flows by estimating the residence times of the different components that contribute to solute transfer between the injection and recovery points. [3] A distributed approach integrating a coupling of geophysical data and pseudo-genetic simulations makes it possible to simulate drainage networks. The associated solute flows and mass transport are then simulated and compared with field data. This is an approach to deduce the structuring of underground flows from the data of artificial tracer tests

Der Zustrom von Grundwasser zu Seen (lacustrine groundwater discharge, LGD) kann signifikante Auswirkungen auf Qualität und Quantität des Seewassers haben. Viele Ansätze zur Identifikation und Quantifizierung von LGD basieren auf Temperaturunterschieden zwischen Grund- und Seewasser und der Messung des damit einhergehenden Wärmetransports. Ziel der Doktorarbeit ist es, Signalfortpflanzung und -ausbreitung des Grundwasserzustroms von der Punktskala an der Sediment-Wasser-Grenzfläche über den Wasserkörper bis zur Grenzfläche Wasseroberfläche-Atmosphäre zu untersuchen. Getestet wird die Hypothese, dass das im Verhältnis zum Umgebungswasser wärmere und daher leichtere Grundwasser in der kalten Wassersäule aufsteigt und die Detektion von LGD an der Wasseroberfläche mit thermalen Infrarot Aufnahmen (TIR) erlaubt. Zunächst wird mit der hierarchical patch dynamics ein Konzept entwickelt, das eine angemessene Kombination multipler Techniken zur Erfassung von Wärme- und Wasserflüssen anbietet (Kap. 2). Dabei werden verschiedene räumliche Skalen und ökohydrologische Grenzflächen abgedeckt. Darauf basierend werden in einem Mesokosmos-Experiment unterschiedliche LGD-Raten durch den Zustrom von warmem Wasser am Grund eines Outdoor-Pools simuliert (Kap. 3 und 4). Ein Glasfaserkabel (fibre-optic distributed temperature sensing, FO-DTS) wird in verschiedenen Tiefen installiert, um das Wärmesignal des Grundwasserzustroms unter verschiedenen Bedingungen zu verfolgen. Mit einer TIR-Kamera wird die Temperatur des Oberflächenwassers aufgezeichnet. Die Aufnahmen werden mit FO-DTS-Temperaturen von 2 cm unter der Wasseroberfläche validiert. Die Anwendung von TIR und FO-DTS ermöglicht die Detektion von LGD in der Wassersäule und an der Grenzfläche Wasseroberfläche-Atmosphäre. Wolkenbedeckung und der Tagesgang der Netto-Strahlung kontrollieren den Auftrieb von LGD und die Zuverlässigkeit der TIR-Ergebnisse. Die besten Ergebnisse werden bei Bewölkung und nachts erzielt.
Lacustrine groundwater discharge (LGD) can have significant impacts on lake water quantity and quality. There is a need to understand LGD mechanisms and to improve measurement methods for LGD. Approaches to identify and quantify LGD are based on significant temperature differences between GW and lake water. The main goal of this PhD thesis is to trace heat signal propagation of LGD from the point scale at the sediment-water interface across the overlying water body to the water surface-atmosphere interface. The PhD thesis tests the hypothesis that the positive buoyancy of warm GW causes upwelling across the cold water column and allows detection of LGD at the water surface by thermal infrared imaging (TIR). First, a general conceptual framework is developed based on hierarchical patch dynamics (HPD). It guides researchers on adequately combining multiple heat tracing techniques to identify and quantify heat and water exchange over several spatial scales and ecohydrological interfaces (Chapter 2). Second, the conceptual framework is used for the design of a mesocosm experiment (Chapters 3 and 4). Different LGD rates were simulated by injecting relatively warm water at the bottom of an outdoor mesocosm. A fiber optic distributed temperature sensing (FO-DTS) cable was installed in a 3D setup in the water column to trace the heat signal of the simulated LGD under different weather conditions and over entire diurnal cycles. Finally, a TIR camera was mounted 4 meters above the mesocosm to monitor water surface temperatures. TIR images were validated using FO-DTS temperature data 2 cm below the water surface (Chapter 4). The positive buoyancy of relatively warm LGD allows the detection of GW across the water column and at the water surface-atmosphere interface by FO-DTS and TIR. Cloud cover and diurnal cycle of net radiation strongly control the upwelling of simulated LGD and the reliability of TIR for detection of LGD at the water surface-atmosphere interface. Optimal results are obtained under overcast conditions and during night.

We are in the era of multicore machines, where we must exploit thread level parallelism for programs to run better, smarter, faster, and more efficiently. In order to increase instruction level parallelism, processors and compilers perform heavy dataflow analyses between instructions. However, there isn’t much work done in the area of inter-thread dataflow analysis. In order to pave the way and find new ways to conserve resources across a variety of domains (i.e., execution speed, chip die area, power efficiency, and computational throughput), we propose a novel framework, termed Contech, to facilitate the analysis of multithreaded program in terms of its communication and execution patterns. We focus the scope on shared memory programs rather than message passing programs, since it is more difficult to analyze the communication and execution patterns for these programs. Discovering patterns of shared memory programs has the potential to allow general purpose computing machines to turn on or off architectural tricks according to application-specific features. Our design of Contech is modular in nature, so we can glean a large variety of information from an architecturally independent representation of the program under examination.

Dans le domaine du calcul de haute performance, les architectures d'exécution sont en continuelle évolution. L'augmentation du nombre de nœuds de calcul, ou le choix d'une topologie réseau plus rapide représentent un investissement important tant en temps qu'en moyen financier. Les méthodes de prédiction de performances permettent de guider ce choix. En parallèle à ce développement, les systèmes HPC pair-à-pair (P2P) se sont également développés ces dernières années. Ce type d'architecture hétérogène permettrait la résolution des problèmes scientifiques pour un coût très faible par rapport au coût d'une architecture dédiée.Ce manuscrit présente une méthode nouvelle de prédiction de performances pour les applications réelles de calcul distribué, exécutées dans des conditions réelles. La prédiction prend en compte l'optimisation du compilateur. Les résultats sont extrapolables et ils sont obtenus pour un ralentissement réduit. Ce travail de recherche est implémenté dans un logiciel nouveau nommé dPerf. dPerf est capable de prédire les performances des applications C, C++ ou Fortran qui communiquent en utilisant les normes MPI ou P2P-SAP et qui s'exécutent sur une architecture cible pair à pair, hétérogène et décentralisée. La précision de cette contribution a été étudiée sur (i) la transformée Laplace, pour l'aspect séquentiel, (ii) le benchmark IS de NAS, pour l'aspect MPI, (iii) et le code de l'obstacle pour l'aspect calcul P2P décentralisé et l'extrapolation du nombre de nœuds.

Dans le domaine du calcul de haute performance, les architectures d’exécution sont en continuelle évolution. L’augmentation du nombre de nœuds de calcul, ou le choix d’une topologie réseau plus rapide représentent un investissement important tant en temps qu’en moyen financier. Les méthodes de prédiction de performances permettent de guider ce choix. En parallèle à ce développement, les systèmes HPC pair-à-pair (P2P) se sont également développés ces dernières années. Ce type d’architecture hétérogène permettrait la résolution des problèmes scientifiques pour un coût très faible par rapport au coût d’une architecture dédiée.Ce manuscrit présente une méthode nouvelle de prédiction de performances pour les applications réelles de calcul distribué, exécutées dans des conditions réelles. La prédiction prend en compte l’optimisation du compilateur. Les résultats sont extrapolables et ils sont obtenus pour un ralentissement réduit. Ce travail de recherche est implémenté dans un logiciel nouveau nommé dPerf. dPerf est capable de prédire les performances des applications C, C++ ou Fortran qui communiquent en utilisant les normes MPI ou P2P-SAP et qui s’exécutent sur une architecture cible pair à pair, hétérogène et décentralisée. La précision de cette contribution a été étudiée sur (i) la transformée Laplace, pour l’aspect séquentiel, (ii) le benchmark IS de NAS, pour l’aspect MPI, (iii) et le code de l’obstacle pour l’aspect calcul P2P décentralisé et l’extrapolation du nombre de nœuds
In the field of high performance computing, the architectures evolve continuously. In order to increase the number of computing nodes or the network speed, an important investment must be considered, from both temporal and financial point of view. Performance prediction methods aim at assisting in finding the best trade-off for such an investment. At the same time, P2P HPC systems have known an increase in development. These heterogeneous architectures would allow solving scientific problems at a low cost, with respect to dedicated systems.The manuscript presents a new method for performance prediction. This method applies to real applications for distributed computing, considered in a real execution environment. This method uses information about the different compiler optimization levels. The prediction results are obtained with reduced slowdown and are scalable. This thesis took shape in the development of the dPerf tool. dPerf predicts the performances of C, C++, and Fortran application, which use MPI or P2P-SAP to communicate. The applications modeled by dPerf are meant for execution on P2P heterogeneous architectures, with a decentralized communication topology. The accuracy of dPerf has been studied on three applications: (i) the Laplace transform, for sequential codes, (ii) the NAS Integer Sort benchmark for distributed MPI programs, (iii) and the obstacle problem, for the decentralized P2P computing and the scaling of the number of computing nodes

Мета роботи: підвищити ефективність управління мережею Microgrid за рахунок вдосконалення системи управління електроенергією сонячних батарей для роботи в режимі реального часу. В даній роботі наведено огляд розумних електромереж, їх роль та взаємодія з існуючими електромережами. Розглядаються різні методи та системи управління мережами MicroGrid, проводиться їх аналіз, характеристика та порівняння, надається характеристика основних рівнів керування мережею MicroGrid. Надається аналіз основних переваг управління сонячною енергією відносно інших джерел енергії в мережі MicroGrid. Досліджується алгоритм визначення положення сонця та враховується вплив погодних умов в реальному часі. Показано, що даний метод підвищує ефективність системи накопичення енергії порівняно з традиційними системами управління енергією.
Goal: improve the efficiency of Microgrid network management by improving the solar power management system for real time operation. This paper provides an overview of smart grids, their role and interaction with existing grids. It discusses the various methods and systems for managing the MicroGrid network, analyzes, characterizes, and compares them, and describes the main layers of MicroGrid network management. The main advantages of solar energy management over other sources of energy in the MicroGrid network are analyzed. The algorithm for determining the position of the sun is investigated and the influence of real-time weather conditions is taken into account. It is shown that this method increases the efficiency of the energy storage system in comparison with traditional energy management systems.

L'objectif de ce travail est d'offrir une aide a la mise au point des applications paralleles et reparties, a base dobjets persistants, permettant une mise au point cyclique et offrant une observation de l'execution dun haut niveau dabstraction. Le non-determinisme et la sensibilite a toute perturbation de ce type d'execution rendent tres difficile la correction des erreurs liees aux conditions d'execution. Les limitations de l'analyse statique des programmes et des approches dynamiques fondees sur une execution courante nous conduisent a preconiser la mise en oeuvre de methodes basees sur la reproduction d'une execution qui apportent une solution au non-determinisme en fixant une execution. La mise au point s'effectue alors dans un contexte particulier ou le comportement de l'execution a corriger est deja connu et peut etre observe a l'aide de vues de l'execution adaptees aux particularites de l'environnement dexecution. Nous definissons, dans le contexte des systemes a objets, un systeme de mise au point base sur la reproduction (dirigee par le controle) d'une execution, permettant une mise au point cyclique et une observation de l'execution au niveau des objets. Nous specifions le service de reexecution, le service d'observation, et proposons une architecture modulaire pour l'assemblage des composants logiciels realisant ces services. Nous presentons ensuite l'application concrete des propositions precedentes au systeme Guide. Nous avons realise un noyau de reexecution, structure en objets Guide, qui se charge de maniere automatique de l'enregistrement et de la reproduction dune execution Guide.

This master's thesis deals with analysis and implementation of methods for BitTorrent monitoring focusing on the Mainline DHT protocol. The aim of the thesis was to create a system, that will be looking for BitTorrent peers that participe in the illegal file distribution. Another task of the system was to collect and analyze data for counting size of the BitTorrent network. That was achieved by taking over of existing method. The system was designed and implemented as a module for monitoring of cybernetic crime. It also defines an interface for storing and sharing data, that provides data evaluation, easy data manipulation and serves for possible future extensions.

Large-scale distributed systems---such as supercomputers, cloud computing platforms, and distributed applications---routinely suffer from slowdowns and crashes due to software and hardware problems, resulting in reduced efficiency and wasted resources. These large-scale systems typically deploy monitoring or tracing systems that gather a variety of statistics about the state of the hardware and the software. State-of-the-art methods either analyze this data manually, or design unique automated methods for each specific problem. This thesis builds on the vision that generalized automated analytics methods on the data sets collected from these complex computing systems provide critical information about the causes of the problems, and this analysis can then enable proactive management to improve performance, resilience, efficiency, or security significantly beyond current limits. This thesis seeks to design scalable, automated analytics methods and frameworks for large-scale distributed systems that minimize dependency on expert knowledge, automate parts of the solution process, and help make systems more resilient. In addition to analyzing data that is already collected from systems, our frameworks also identify what to collect from where in the system, such that the collected data would be concise and useful for manual analytics. We focus on two data sources for conducting analytics: numeric telemetry data, which is typically collected from operating system or hardware counters, and end-to-end traces collected from distributed applications. This thesis makes the following contributions in large-scale distributed systems: (1) Designing a framework for accurately diagnosing previously encountered performance variations, (2) designing a technique for detecting (unwanted) applications running on the systems, (3) developing a suite for reproducing performance variations that can be used to systematically develop analytics methods, (4) designing a method to explain predictions of black-box machine learning frameworks, and (5) constructing an end-to-end tracing framework that can dynamically adjust instrumentation for effective diagnosis of performance problems.
2021-09-28T00:00:00Z

碩士
國立臺灣大學
職業醫學與工業衛生研究所
97
The water supply system in Taiwan is quite well established and its serving rate is extremely high. However, recent studies have indicated that adverse effects resulting from exposure to trace metals in tap water could occur at low levels. Therefore, the main purpose of this study were to characterize the distribution of trace metal levels in raw, finished, distributed water, in order to explore the source of and the effect of exposure to trace metals in water, and to correlate the trace metal levels in distributed water and cord blood. Trace metal levels in water of 92 domestic water treatment plants were analyzed in this study. Inductively coupled plasma - mass spectrometry (ICP-MS) was used for the analysis of 20 trace metals in raw, finished and distributed water. Furthermore, the results were examined for in the association with geologic properties, water treatment processes and trace metals in cord blood. Geographic information system (GIS) was utilized to identify the location of, to localize the buffer zones of and to overlap layers on GIS for water treatment plants. Analytical data was integrated and analyzed by Statistical Analysis System (SAS). Results showed that the mean arsenic concentration was 1.49±1.51 µg/L in raw water from the alluvium area and was higher than those from other geologic area (p<0.05). Mean arsenic concentrations in raw water from Yunlin and Chiayi, both located in the alluvium area, were 3.12±1.07 µg/L and 2.87±0.11 µg/L, respectively, relatively higher than those in other regions (p<0.05). Certain trace metals, including As, Be, Co, Cr, Mn, Ni, Tl, V increased levels in finished water after water treatment while others decreased, i.e., Sb, Se, Ag. Within water serving zones with radius of 2, 3 or 4 kilometers, positive correlations were found for trace metals of Be, Cd, U, Mo, Sb between distributed water and cord blood. In summary, these findings in the present study will be helpful for relevant studies in future to further explore the association between trace metals background levels in raw water and the health potential effects in human.

This thesis evaluates a method for extracting architectural dependencies and performance measures from an evolving distributed software system. The research goal was to establish methods of determining potential scalability issues in a distributed software system as it is being iteratively developed. The research evaluated the use of industry available distributed tracing methods to extract performance measures and queuing network model parameters for common user activities. Additionally, a method was developed to trace and collect system operations the correspond to these user activities utilizing automated acceptance testing. Performance measure extraction was tested across several historical releases of a real-world distributed software system with this method. The trends in performance measures across releases correspond to several scalability issues identified in the production software system.
Graduate

In recent years, wireline temperature profiling methods have evolved to offer new insight into fractured rock hydrogeology. An important advance in temperature logging makes use of boreholes temporarily sealed with flexible impervious fabric liners so that the water column is static and effects of cross-connection are eliminated. For this project a characterization technique was developed based on combining fibre optic distributed temperatures sensing (DTS) with active heating within boreholes sealed with underground liners. DTS systems provide a temperature profiling method that offers improved temporal resolution when compared with wireline trolling based techniques. The ability to collect temperature profiles rapidly in time can improve understanding of transient processes. In this study the advantage of a sealed borehole environment for temperature investigations is demonstrated. Evidence for identifying active groundwater flow under natural gradient conditions using DTS heat pulse testing is presented through a comparison with high resolution geologic logging and hydraulic datasets.

Prague and Brno belong to one of the most contaminated cities of Czech Republic. The main objective of this study was to compare PGE contents and concentrations of risk elements (Zn, Cd, Pb, Cu, As, Sb, Hg) in the soils of both cities's municipal parks. Soil samples were taken from depths of 0-10 cm, 10-20 cm and 20-30 cm. Besides soil samples, the sand samples from children sandpits were taken from depth 0-30 cm. In 52 samples of soil and sand Pd, Pt and Rh contents were determinated by ICP- MS after Ni-S fire assay procedure. Soil samples were additionally analysed on Pb isotopic composition. Higher PGE contents were determined in Prague; the highest contents were measured at location of Ortenovo náměstí (50,5 μg·kg-1 Pt, 33,9 μg·kg-1 Pd a 11,3 μg·kg-1 Rh). Elevated PGE concentrations were found in the upper layers of most soils in both cities. Elevated contents of risk elements were determined at locations with high traffic density. Most of samples showed that contents of risk elements decrease with depth. The highest concentrations reached Zn (394 mg·kg-1 ), Pb (290 mg·kg-1 ) and Cu (181 mg·kg-1 ). In Prague, the isotopic ratios 206 Pb/207 Pb vary from 1,136 to 1,181; in Brno from 1,161 to 1,192. Measured isotopic ratios suggest contamination both from gasoline and ore combustion. Key words:...

Carbonatites are unique and enigmatic magmatic rocks of unclear origin, with very specific mineralogy and geochemical properties. They are predominantly composed of magmatic calcite or other carbonate minerals (Le Bas 1987) and have low content of SiO2 (Le Maitre 2002). Origin of these peculiar magmas is still not clear but they appear to represent an important "window" into processes in Earth's mantle. They are considered either as residual melts from a fractionated carbonated nephelinite or melilitite (Gittins 1989; Gittins and Jago 1998), as immiscible fractions of CO2-saturated silicate melts (Freestone and Hamilton 1980; Amundsen 1987; Kjarsgaard and Hamilton 1988, 1989; Brooker and Hamilton 1990; Kjarsgaard and Peterson 1991; Church and Jones 1995; Lee and Wyllie 1997; Dawson 1998; Halama et al. 2005; Brooker and Kjarsgaard 2011), or as primary melts which are were generated from CO2-bearing peridotite through partial melting (Wallace and Green 1988; Sweeney 1994; Harmer and Gittins 1998; Harmer et al. 1998; Ying et al. 2004). Abundances of rare earth elements (REE) are often high in carbonatites because carbonatitic magmas can dissolve these elements much easily than silicate magmas (Nelson et al. 1988). Carbonatitic magma can also dissolve large quantities of Sr, Ba, P and mainly Zr and Nb,...

The thesis examined the comprehensive causal relationship between the banking sector, stock market development and economic growth in a multi-variate framework using Zimbabwean time series data from 1988 to 2015. Three banking sector development proxies (total financial sector credit, banking credit to private sector and broad money M3) and three stock market development proxies (stock market capitalization, value traded and turnover ratio) were employed to estimate both long and short run relationships between banking sector, stock market and economic growth in Zimbabwe. The study employs the vector error correction model (VECM) as the main estimation technique and the autoregressive distributed lag (ARDL) approach as a robustness testing technique. Results showed that in Zimbabwe a significant causal relationship from banking sector and stock market development to economic growth exists in the long run without any feedback effects. In the short run, however, a negative yet statistically significant causal relationship runs from economic growth to banking sector and stock market development in Zimbabwe. The study further concludes that there is a unidirectional causal relationship running from stock market development to banking sector development in Zimbabwe in both short and long run periods. Nonetheless this relationship between banking sector and stock markets has been found to be more significant in the short run than in the long run. The thesis adopts the complementary view and recommends for the spontaneity implementation of monetary policies as the economy grows. Monetary authorities should thus formulate policies to promote both banks and stock markets with corresponding growth in Zimbabwe’s economy.
Business Management
M. Com. (Business Management)