A considerably fraction of science discovery is nowadays relying on computer simulations. High Performance Computing  (HPC) provides scientists with the means to simulate processes ranging from climate modeling to protein folding. However, achieving good application performance and making an optimal use of HPC resources is a heroic task due to the complexity of parallel software. Therefore, performance tools  and runtime systems that help users to execute  applications in the most optimal way are of utmost importance in the landscape of HPC.  In this thesis, we explore different techniques to tackle the challenges of collecting, storing, and using  fine-grained performance data. First, we investigate the automatic use of real-time performance data in order to run applications in an optimal way. To that end, we present a prototype of an adaptive task-based runtime system that uses real-time performance data for task scheduling. This runtime system has a performance monitoring component that provides real-time access to the performance behavior of anapplication while it runs. The implementation of this monitoring component is presented and evaluated within this thesis. Secondly, we explore lossless compression approaches  for MPI monitoring. One of the main problems that  performance tools face is the huge amount of fine-grained data that can be generated from an instrumented application. Collecting fine-grained data from a program is the best method to uncover the root causes of performance bottlenecks, however, it is unfeasible with extremely parallel applications  or applications with long execution times. On the other hand, collecting coarse-grained data is scalable but  sometimes not enough to discern the root cause of a performance problem. Thus, we propose a new method for performance monitoring of MPI programs using event flow graphs. Event flow graphs  provide very low overhead in terms of execution time and  storage size, and can be used to reconstruct fine-grained trace files of application events ordered in time. / <p>QC 20150508</p>

L'apparition des Systèmes d'Information ubiquitaires ou pervasifs est issue de l'émergence de nouvelles technologies fournissant au système une vision de son environnement, de l'environnement de ses utilisateurs ainsi que de leurs profils. Grâce à ces données formant le contexte de l'application, il est possible de fournir des services personnalisés, pertinents et ciblés. Mais, le problème qui se pose à ce niveau concerne le degré d'adaptation, de prise de décision à la place de l'utilisateur et de l'identification des données contextuelles nécessaires et suffisantes pour ces services. Ceci est dû à un déséquilibre entre les avancées des technologies et de leurs applications (qui reçoivent un grand intérêt de la part de la recherche et de l'industrie) et les méthodes et démarches de développement et d'ingénierie spécifiques aux systèmes ubiquitaires. Notre objectif dans ce travail de thèse est de proposer une méthode d'ingénierie des SI ubiquitaires en considérant les différentes exigences reliées à la nature mobile et grande échelle de ces systèmes. Cette méthode est basée sur une démarche de développement qui fait usage d'un ensemble de métamodèles et de langages génériques favorisant la spécification complète de ces systèmes. Cette démarche sépare les spécifications fonctionnelles, techniques et ubiquitaires. Les spécifications ubiquitaires permettent de définir des modèles structurels et événementiels du contexte respectant les exigences des utilisateurs et les contraintes de sécurité et supportant les fonctionnalités d'adaptation et de sensibilité au contexte. Cette approche orientée évènements est consolidée par l'adoption d'une architecture de traitement des évènements. Notre démarche E-CARe est une intégration des spécifications ubiquitaires dans une démarche classique de conception des SI pour garantir la couverture des spécifications fonctionnelles et techniques. Les applications d'assistance représentent un cas d'étude idéal pour cette démarche qui s'intéresse au domaine des transports, fortement dépendant de l'environnement et des évènements ambiants. / Ubiquitous Information Systems appeared as a consequence to emerging and evolving communication and mobile technologies providing the system with information on its environment, the environment of its users and their profiles. These data constitute the application context and are used to provide personalized, targeted and relevant services. However, ubiquitous services face some difficulties and challenges concerning specially needed contextual data, adaptation degree and computerized decision making. This is due to the gap between advanced ubiquitous services and their applications, and methods and processes for developing and engineering ubiquitous systems. Our goal in this thesis is to propose an engineering method for ubiquitous Information Systems considering different requirements resulting from the mobile and high scalable nature of these systems. The proposed method is based on a development process and a set of generic metamodels and languages facilitating a complete system specification and implementation. The proposed process separates functional, technical and ubiquitous specifications. Ubiquitous specifications enable the structural and event based context models definition while considering user requirements and security requirements. Adaptation and context awareness functionalities are supported by structural and dynamic context models. The proposed event oriented approach is enhanced by the adoption of an event processing architecture. Ubiquitous specifications are integrated into a classical information systems engineering process to constitute the E-CARe process including functional and technical specifications. Our propositions are used to design a user assistance application in the transport domain, highly dependent on the ambient environment and events.