Les systèmes multiprocesseurs sur puce, tel que le processeur CELL ou plus récemment Platform 2012, sont des architectures multicœurs hétérogènes constitués d'un processeur host et d'une fabric de calcul qui consiste en plusieurs petits cœurs dont le rôle est d'agir comme un accélérateur programmable. Les parties parallélisable d'une application, qui initialement est supposé etre executé par le host, et dont le calcul est intensif sont envoyés a la fabric multicœurs pour être exécutés. Ces applications sont en général des applications qui manipulent des tableaux trés larges de données, ces données sont stockées dans une memoire distante hors puce (off-chip memory) dont l 'accès est 100 fois plus lent que l 'accès par un cœur a une mémoire locale. Accéder ces données dans la mémoire off-chip devient donc un problème majeur pour les performances. une characteristiques principale de ces plateformes est une mémoire local géré par le software, au lieu d un mechanisme de cache, tel que les mouvements de données dans la hiérarchie mémoire sont explicitement gérés par le software. Dans cette thèse, l 'objectif est d'optimiser ces transfert de données dans le but de reduire/cacher la latence de la mémoire off-chip . / Multiprocessor system on chip (MPSoC) such as the CELL processor or the more recent Platform2012 are heterogeneous multi-core architectures, with a powerful host processor and a computation fabric, consisting of several smaller cores, whose intended role is to act as a general purpose programmable accelerator. Therefore computation-intensive (and parallelizable) parts of the application initially intended to be executed by the host processor are offloaded to the multi-cores for execution. These parts of the application are often data intensive, operating on large arrays of data initially stored in a remote off-chip memory whose access time is about 100 times slower than that of the cores local memory. Accessing data in the off-chip memory becomes then a main bottleneck for performance. A major characteristic of these platforms is a software controlled local memory storage rather than a hidden cache mechanism where data movement in the memory hierarchy, typically performed using a DMA (Direct Memory Access) engine, are explicitely managed by the software. In this thesis, we attempt to optimize such data transfers in order to reduce/hide the off-chip memory latency.

Au cours des dernières années, l'informatique “green” est devenue un sujet important dans le calcul intensif. Cependant, les plates-formes informatiques continuent de consommer de plus en plus d'énergie en raison de l'augmentation du nombre de noeuds qui les composent. Afin de minimiser les coûts d'exploitation de ces plates-formes de nombreuses techniques ont été étudiées, parmi celles-ci, il y a le changement de la fréquence dynamique des processeurs (DVFS en anglais). Il permet de réduire la consommation d'énergie d'un CPU, en abaissant sa fréquence. Cependant, cela augmente le temps d'exécution de l'application. Par conséquent, il faut trouver un seuil qui donne le meilleur compromis entre la consommation d'énergie et la performance d'une application. Cette thèse présente des algorithmes développés pour optimiser la consommation d'énergie et les performances des applications parallèles avec des itérations synchrones et asynchrones sur des clusters ou des grilles. Les modèles de consommation d'énergie et de performance proposés pour chaque type d'application parallèle permettent de prédire le temps d'exécution et la consommation d'énergie d'une application pour toutes les fréquences disponibles.La contribution de cette thèse peut être divisé en trois parties. Tout d'abord, il s'agit d'optimiser le compromis entre la consommation d'énergie et les performances des applications parallèles avec des itérations synchrones sur des clusters homogènes. Deuxièmement, nous avons adapté les modèles de performance énergétique aux plates-formes hétérogènes dans lesquelles chaque noeud peut avoir des spécifications différentes telles que la puissance de calcul, la consommation d'énergie, différentes fréquences de fonctionnement ou encore des latences et des bandes passantes réseaux différentes. L'algorithme d'optimisation de la fréquence CPU a également été modifié en fonction de l'hétérogénéité de la plate-forme. Troisièmement, les modèles et l'algorithme d'optimisation de la fréquence CPU ont été complètement repensés pour prendre en considération les spécificités des algorithmes itératifs asynchrones.Tous ces modèles et algorithmes ont été appliqués sur des applications parallèles utilisant la bibliothèque MPI et ont été exécutés avec le simulateur Simgrid ou sur la plate-forme Grid'5000. Les expériences ont montré que les algorithmes proposés sont plus efficaces que les méthodes existantes. Ils n’introduisent qu’un faible surcoût et ne nécessitent pas de profilage au préalable car ils sont exécutés au cours du déroulement de l’application. / In recent years, green computing has become an important topic in the supercomputing research domain. However, the computing platforms are still consuming more and more energy due to the increase in the number of nodes composing them. To minimize the operating costs of these platforms many techniques have been used. Dynamic voltage and frequency scaling (DVFS) is one of them. It can be used to reduce the power consumption of the CPU while computing, by lowering its frequency. However, lowering the frequency of a CPU may increase the execution time of the application running on that processor. Therefore, the frequency that gives the best trade-off between the energy consumption and the performance of an application must be selected.This thesis, presents the algorithms developed to optimize the energy consumption and theperformance of synchronous and asynchronous message passing applications with iterations runningover clusters or grids. The energy consumption and performance models for each type of parallelapplication predicts its execution time and energy consumption for any selected frequency accordingto the characteristics of both the application and the architecture executing this application.The contribution of this thesis can be divided into three parts: Firstly, optimizing the trade-offbetween the energy consumption and the performance of the message passing applications withsynchronous iterations running over homogeneous clusters. Secondly, adapting the energy andperformance models to heterogeneous platforms where each node can have different specificationssuch as computing power, energy consumption, available frequency gears or network’s latency andbandwidth. The frequency scaling algorithm was also modified to suit the heterogeneity of theplatform. Thirdly, the models and the frequency scaling algorithm were completely rethought to takeinto considerations the asynchronism in the communication and computation. All these models andalgorithms were applied to message passing applications with iterations and evaluated over eitherSimGrid simulator or Grid’5000 platform. The experiments showed that the proposed algorithms areefficient and outperform existing methods such as the energy and delay product. They also introducea small runtime overhead and work online without any training or profiling.

Les systèmes distribués, tels que les grilles, sont utilisés aujourd’hui pour l’exécution des grandes applications parallèles. Quelques caractéristiques de ces systèmes sont l’interconnexion complexe de ressources qui pourraient être présent et de la facile passage à l’échelle. La complexité d’interconnexion vient, par exemple, d’un nombre plus grand de directives de routage pour la communication entre les processus et une latence variable dans le temps. La passage à l’échelle signifie que des ressources peuvent être ajoutées indéfiniment simplement en les reliant à l’infrastructure existante. Ces caractéristiques influencent directement la façon dont la performance des applications parallèles doit être analysée. Les techniques de visualisation traditionnelles pour cette analyse sont généralement basées sur des diagrammes de Gantt que disposent la liste des composants de l’application verticalement et metent la ligne du temps sur l’axe horizontal. Ces représentations visuelles ne sont généralement pas adaptés à l’analyse des applications exécutées en parallèle dans les grilles. La première raison est qu’elles n’ont pas été conçues pour offrir aux développeurs une analyse qui montre aussi la topologie du réseau des ressources. La deuxième raison est que les techniques de visualisation traditionnels ne s’adaptent pas bien quand des milliers d’entités doivent être analysés ensemble. Cette thèse tente de résoudre les problèmes des techniques traditionnelles dans la visualisation des applications parallèles. L’idée principale est d’exploiter le domaine de la visualisation de l’information et essayer d’appliquer ses concepts dans le cadre de l’analyse des programmes parallèles. Portant de cette idée, la thèse propose deux modèles de visualisation : les trois dimensions et le modèle d’agrégation visuelle. Le premier peut être utilisé pour analyser les programmes parallèles en tenant compte de la topologie du réseau. L’affichage lui-même se compose de trois dimensions, où deux sont utilisés pour indiquer la topologie et la troisième est utilisée pour représenter le temps. Le second modèle peut être utilisé pour analyser des applications parallèles comportant un très grand nombre de processsus. Ce deuxième modèle exploite une organisation hiérarchique des données utilisée par une technique appelée Treemap pour représenter visuellement la hiérarchie. Les implications de cette thèse sont directement liées à l’analyse et la compréhension des applications parallèles exécutés dans les systèmes distribués. Elle améliore la compréhension des modes de communication entre les processus et améliore la possibilité d’assortir les motifs avec cette topologie de réseau réel sur des grilles. Bien que nous utilisons abondamment l’exemple de la topologie du réseau, l’approche pourrait être adapté, avec presque pas de changements, à l’interconnexion fourni par un middleware d’une interconnexion logique. Avec la technique d’agrégation, les développeurs sont en mesure de rechercher des patterns et d’observer le comportement des applications à grande échelle. / Sistemas distribuídos tais como grids são usados hoje para a execução de aplicações paralelas com um grande número de processos. Algumas características desses sistemas são a presença de uma complexa rede de interconexão e a escalabilidade de recursos. A complexidade de rede vem, por exemplo, de largura de banda e latências variáveis ao longo do tempo. Escalabilidade é a característica pela qual novos recursos podem ser adicionados em um grid apenas através da conexão em uma infraestrutura pré-existente. Estas características influenciam a forma como o desempenho de aplicações paralelas deve ser analisado. Esquemas tradicionais de visualização de desempenho são usualmente baseados em gráficos Gantt com uma dimensão para listar entidades monitoradas e outra para o tempo. Visualizações como essa não são apropriadas para a análise de aplicações paralelas executadas em grid. A primeira razão para tal é que elas não foram concebidas para oferecer ao desenvolvedor uma análise que mostra a topologia dos recursos e a relação disso com a aplicação. A segunda razão é que técnicas tradicionais não são escaláveis quando milhares de entidades monitoradas devem ser analisadas conjuntamente. Esta tese tenta resolver estes problemas encontrados em técnicas de visualização tradicionais para a análise de aplicações paralelas. A idéia principal consiste em explorar técnicas da área de visualização da informação e aplicá-las no contexto de análise de programas paralelos. Levando em conta isto, esta tese propõe dois modelos de visualização: o de três dimensões e o modelo de agregação visual. O primeiro pode ser utilizado para analisar aplicações levando-se em conta a topologia da rede dos recursos. A visualização em si é composta por três dimensões, onde duas são usadas para mostrar a topologia e a terceira é usada para representar o tempo. O segundo modelo pode ser usado para analisar aplicações paralelas com uma grande quantidade de processos. Ela explora uma organização hierárquica dos dados de monitoramento e uma técnica de visualização chamada Treemap para representar visualmente a hierarquia. Os dois modelos representam uma nova forma de analisar aplicação paralelas visualmente, uma vez que eles foram concebidos para larga-escala e sistemas distribuídos complexos, como grids. As implicações desta tese estão diretamente relacionadas à análise e ao entendimento do comportamento de aplicações paralelas executadas em sistemas distribuídos. Um dos modelos de visualização apresentados aumenta a compreensão dos padrões de comunicação entre processos e oferece a possibilidade de observar tal padrão com a topologia de rede. Embora a topologia de rede seja usada, a abordagem pode ser adaptada sem grandes mudanças para levar em conta interconexões lógicas de bibliotecas de comunicação. Com a técnica de agregação apresentada nesta tese, os desenvolvedores são capazes de observar padrões de aplicações paralelas de larga escala. / Highly distributed systems such as grids are used today for the execution of large-scale parallel applications. Some characteristics of these systems are the complex resource interconnection that might be present and the scalability. The interconnection complexity comes from the different number of hops to provide communication among applications processes and differences in network latencies and bandwidth. The scalability means that the resources can be added indefinitely just by connecting them to the existing infrastructure. These characteristics influence directly the way parallel applications performance must be analyzed. Current traditional visualization schemes to this analysis are usually based on Gantt charts with one dimension to list the monitored entities and the other dimension dedicated to time. These visualizations are generally not suited to parallel applications executed in grids. The first reason is that they were not built to offer to the developer an analysis that also shows the network topology of the resources. The second reason is that traditional visualization techniques do not scale well when thousands of monitored entities must be analyzed together. This thesis tries to overcome the issues encountered on traditional visualization techniques for parallel applications. The main idea behind our efforts is to explore techniques from the information visualization research area and to apply them in the context of parallel applications analysis. Based on this main idea, the thesis proposes two visualization models: the three-dimensional and the visual aggregation model. The former might be used to analyze parallel applications taking into account the network topology of the resources. The visualization itself is composed of three dimensions, where two of them are used to render the topology and the third is used to represent time. The later model can be used to analyze parallel applications composed of several thousands of processes. It uses hierarchical organization of monitoring data and an information visualization technique called Treemap to represent that hierarchy. Both models represent a novel way to visualize the behavior of parallel applications, since they are conceived considering large-scale and complex distributed systems, such as grids. The implications of this thesis are directly related to the analysis and understanding of parallel applications executed in distributed systems. It enhances the comprehension of patterns in communication among processes and improves the possibility of matching this patterns with real network topology of grids. Although we extensively use the network topology example, the approach could be adapted with almost no changes to the interconnection provided by a middleware of a logical interconnection. With the scalable visualization technique, developers are able to look for patterns and observe the behavior of large-scale applications.

As computational grids have become popular and ubiquitous, users have access to large number and different types of geographically distributed grid resources. Many computational grid frameworks are composed of multiple distributed sites with each site consisting of one or more dedicated or non-dedicated clusters. Jobs submitted to a grid are handled by a matascheduler which interacts with the local schedulers of the clusters for scheduling jobs to the individual clusters. Computational grids have been found to be powerful research-beds for execution of various kinds of parallel applications. When a parallel application is submitted to a grid, the metascheduler has to choose a set of resources from a cluster for application execution. To select the best set of resources for application execution, it is important to determine the performance of the application. Accurate performance estimates of an application is essential in assisting a grid meta scheduler to efficiently schedule user jobs. Thus models that predict execution times of parallel applications on a set of resources and a search procedure (scheduling strategy) which selects the best set of machines within a cluster for application execution are of importance for enabling the parallel applications on grids. For efficient execution of large scientific parallel applications consisting of multiple phases, performance models of the individual phases should be obtained. Efficient rescheduling strategies that can use the per-phase models to adapt the parallel applications to application and resource dynamics are necessary for maintaining high performance of the applications on grids. A practical and robust grid computing infrastructure that integrates components related to application and resource monitoring, performance modeling, scheduling and rescheduling techniques, is highly essential for large-scale deployment and high performance of scientific applications on grid systems and hence for fostering high performance computing. This thesis focuses on developing performance models for predicting execution times of parallel problems/subproblems on dedicated and non-dedicated grid resources. The thesis also constructs robust scheduling and rescheduling strategies in a grid metascheduler that can use the performance models for efficient execution of large scientific parallel applications on dynamic grids. Finally, the thesis builds a practical and robust grid middleware infrastructure which integrates components related to performance modeling, scheduling and rescheduling, monitoring and migration frameworks for large-scale deployment and use of high performance applications on grids. The thesis consists of four main components. In the first part of the thesis, we have developed a comprehensive set of performance modeling strategies to predict the execution times of tightly-coupled parallel applications on a set of resources in a dedicated or non-dedicated cluster. The main purpose of our prediction strategies is to aid grid metaschedulers in making scheduling decisions. Our performance modeling strategies, based on linear regression, can deal with non-dedicated systems where the loads can change during application executions. Our models do not require detailed knowledge and instrumentation of the applications and can be constructed without the involvement of application developers. The strategies are intended for rapid and large scale deployment of parallel applications on non-dedicated grid systems. We have evaluated our strategies on 8, 16, 24 and 32-node clusters with random loads and load traces from a grid system. Our performance modeling strategies gave less than 30% average percentage prediction errors in all cases, which is reasonable for non-dedicated systems. We also found that scheduling based on the predictions by our strategies will result in perfect scheduling in many cases. For modeling large-scale scientific applications, we use execution profiles and automatic program analysis, and manual analysis of significant portions of the application’s code to identify the different phases of applications. We then adopt our performance modeling strategies to predict execution times for the different phases of the tightly-coupled parallel applications on a set of resources in a dedicated or non-dedicated cluster. Our experiments show that using combinations of performance models of the phases give 18% – 70% more accurate predictions than using single performance models for the applications. In the second part of the thesis, we have devised, evaluated and compared algorithms for scheduling tightly-coupled parallel applications on multi-cluster grids. Our algorithms use performance models that predict the execution times of parallel applications, for evaluations of candidate schedules. In this work, we propose a novel algorithm called Box Elimination (BE) that searches a space of performance model parameters to determine efficient schedules. By eliminating large search space regions containing poorer solutions at each step and searching high quality solutions, our algorithm is able to generate efficient schedules within few seconds for even clusters of 512 processors. By means of large number of real and simulation experiment, we compared our algorithm with popular optimization techniques. We show that our algorithm generates up to 80% more efficient schedules than other algorithms and the resulting execution times are more robust against performance modeling errors. The third part of the thesis deals with policies for rescheduling long-running multi-phase parallel applications in response to application and resource dynamics. In this work, we use our performance modeling and scheduling strategies to derive rescheduling plans for executing multi-phase parallel applications on grids. A rescheduling plan consists of potential points in application execution for rescheduling and schedules of resources for application execution between two consecutive rescheduling points. We have developed three algorithms, namely an incremental algorithm, a divide-and-conquer algorithm and a genetic algorithm, for deriving a rescheduling plan for a parallel application execution. We have also developed an algorithm that uses rescheduling plans derived on different clusters to form a single coherent rescheduling plan for application execution on a grid consisting of multiple clusters. The rescheduling plans generated by our algorithms are highly efficient leading to application execution times that are higher than the execution times corresponding to brute force method by less than 10%. We also find that rescheduling in response to changing application and resource dynamics, using the rescheduling plans for multi-cluster grids generated by our algorithms, give much lesser execution times when compared to executions of the applications on a single schedule throughout application execution. In the final part of the thesis, we have developed a practical grid middleware framework called MerITA (Middleware for Performance Improvement of Tightly Coupled Parallel Applications on Grids), a system for effective execution of tightly-coupled parallel applications on multi-cluster grids consisting of dedicated or non-dedicated, interactive or batch systems. The framework brings together performance modeling for automatically determining the characteristics of parallel applications, scheduling strategies that use the performance models for efficient mapping of applications to resources, rescheduling policies for determining the points in application execution when executing applications can be rescheduled to different sets of resources to obtain performance improvement and a check-pointing library for enabling rescheduling.

Les systèmes distribués, tels que les grilles, sont utilisés aujourd’hui pour l’exécution des grandes applications parallèles. Quelques caractéristiques de ces systèmes sont l’interconnexion complexe de ressources qui pourraient être présent et de la facile passage à l’échelle. La complexité d’interconnexion vient, par exemple, d’un nombre plus grand de directives de routage pour la communication entre les processus et une latence variable dans le temps. La passage à l’échelle signifie que des ressources peuvent être ajoutées indéfiniment simplement en les reliant à l’infrastructure existante. Ces caractéristiques influencent directement la façon dont la performance des applications parallèles doit être analysée. Les techniques de visualisation traditionnelles pour cette analyse sont généralement basées sur des diagrammes de Gantt que disposent la liste des composants de l’application verticalement et metent la ligne du temps sur l’axe horizontal. Ces représentations visuelles ne sont généralement pas adaptés à l’analyse des applications exécutées en parallèle dans les grilles. La première raison est qu’elles n’ont pas été conçues pour offrir aux développeurs une analyse qui montre aussi la topologie du réseau des ressources. La deuxième raison est que les techniques de visualisation traditionnels ne s’adaptent pas bien quand des milliers d’entités doivent être analysés ensemble. Cette thèse tente de résoudre les problèmes des techniques traditionnelles dans la visualisation des applications parallèles. L’idée principale est d’exploiter le domaine de la visualisation de l’information et essayer d’appliquer ses concepts dans le cadre de l’analyse des programmes parallèles. Portant de cette idée, la thèse propose deux modèles de visualisation : les trois dimensions et le modèle d’agrégation visuelle. Le premier peut être utilisé pour analyser les programmes parallèles en tenant compte de la topologie du réseau. L’affichage lui-même se compose de trois dimensions, où deux sont utilisés pour indiquer la topologie et la troisième est utilisée pour représenter le temps. Le second modèle peut être utilisé pour analyser des applications parallèles comportant un très grand nombre de processsus. Ce deuxième modèle exploite une organisation hiérarchique des données utilisée par une technique appelée Treemap pour représenter visuellement la hiérarchie. Les implications de cette thèse sont directement liées à l’analyse et la compréhension des applications parallèles exécutés dans les systèmes distribués. Elle améliore la compréhension des modes de communication entre les processus et améliore la possibilité d’assortir les motifs avec cette topologie de réseau réel sur des grilles. Bien que nous utilisons abondamment l’exemple de la topologie du réseau, l’approche pourrait être adapté, avec presque pas de changements, à l’interconnexion fourni par un middleware d’une interconnexion logique. Avec la technique d’agrégation, les développeurs sont en mesure de rechercher des patterns et d’observer le comportement des applications à grande échelle. / Sistemas distribuídos tais como grids são usados hoje para a execução de aplicações paralelas com um grande número de processos. Algumas características desses sistemas são a presença de uma complexa rede de interconexão e a escalabilidade de recursos. A complexidade de rede vem, por exemplo, de largura de banda e latências variáveis ao longo do tempo. Escalabilidade é a característica pela qual novos recursos podem ser adicionados em um grid apenas através da conexão em uma infraestrutura pré-existente. Estas características influenciam a forma como o desempenho de aplicações paralelas deve ser analisado. Esquemas tradicionais de visualização de desempenho são usualmente baseados em gráficos Gantt com uma dimensão para listar entidades monitoradas e outra para o tempo. Visualizações como essa não são apropriadas para a análise de aplicações paralelas executadas em grid. A primeira razão para tal é que elas não foram concebidas para oferecer ao desenvolvedor uma análise que mostra a topologia dos recursos e a relação disso com a aplicação. A segunda razão é que técnicas tradicionais não são escaláveis quando milhares de entidades monitoradas devem ser analisadas conjuntamente. Esta tese tenta resolver estes problemas encontrados em técnicas de visualização tradicionais para a análise de aplicações paralelas. A idéia principal consiste em explorar técnicas da área de visualização da informação e aplicá-las no contexto de análise de programas paralelos. Levando em conta isto, esta tese propõe dois modelos de visualização: o de três dimensões e o modelo de agregação visual. O primeiro pode ser utilizado para analisar aplicações levando-se em conta a topologia da rede dos recursos. A visualização em si é composta por três dimensões, onde duas são usadas para mostrar a topologia e a terceira é usada para representar o tempo. O segundo modelo pode ser usado para analisar aplicações paralelas com uma grande quantidade de processos. Ela explora uma organização hierárquica dos dados de monitoramento e uma técnica de visualização chamada Treemap para representar visualmente a hierarquia. Os dois modelos representam uma nova forma de analisar aplicação paralelas visualmente, uma vez que eles foram concebidos para larga-escala e sistemas distribuídos complexos, como grids. As implicações desta tese estão diretamente relacionadas à análise e ao entendimento do comportamento de aplicações paralelas executadas em sistemas distribuídos. Um dos modelos de visualização apresentados aumenta a compreensão dos padrões de comunicação entre processos e oferece a possibilidade de observar tal padrão com a topologia de rede. Embora a topologia de rede seja usada, a abordagem pode ser adaptada sem grandes mudanças para levar em conta interconexões lógicas de bibliotecas de comunicação. Com a técnica de agregação apresentada nesta tese, os desenvolvedores são capazes de observar padrões de aplicações paralelas de larga escala. / Highly distributed systems such as grids are used today for the execution of large-scale parallel applications. Some characteristics of these systems are the complex resource interconnection that might be present and the scalability. The interconnection complexity comes from the different number of hops to provide communication among applications processes and differences in network latencies and bandwidth. The scalability means that the resources can be added indefinitely just by connecting them to the existing infrastructure. These characteristics influence directly the way parallel applications performance must be analyzed. Current traditional visualization schemes to this analysis are usually based on Gantt charts with one dimension to list the monitored entities and the other dimension dedicated to time. These visualizations are generally not suited to parallel applications executed in grids. The first reason is that they were not built to offer to the developer an analysis that also shows the network topology of the resources. The second reason is that traditional visualization techniques do not scale well when thousands of monitored entities must be analyzed together. This thesis tries to overcome the issues encountered on traditional visualization techniques for parallel applications. The main idea behind our efforts is to explore techniques from the information visualization research area and to apply them in the context of parallel applications analysis. Based on this main idea, the thesis proposes two visualization models: the three-dimensional and the visual aggregation model. The former might be used to analyze parallel applications taking into account the network topology of the resources. The visualization itself is composed of three dimensions, where two of them are used to render the topology and the third is used to represent time. The later model can be used to analyze parallel applications composed of several thousands of processes. It uses hierarchical organization of monitoring data and an information visualization technique called Treemap to represent that hierarchy. Both models represent a novel way to visualize the behavior of parallel applications, since they are conceived considering large-scale and complex distributed systems, such as grids. The implications of this thesis are directly related to the analysis and understanding of parallel applications executed in distributed systems. It enhances the comprehension of patterns in communication among processes and improves the possibility of matching this patterns with real network topology of grids. Although we extensively use the network topology example, the approach could be adapted with almost no changes to the interconnection provided by a middleware of a logical interconnection. With the scalable visualization technique, developers are able to look for patterns and observe the behavior of large-scale applications.

Les systèmes distribués, tels que les grilles, sont utilisés aujourd’hui pour l’exécution des grandes applications parallèles. Quelques caractéristiques de ces systèmes sont l’interconnexion complexe de ressources qui pourraient être présent et de la facile passage à l’échelle. La complexité d’interconnexion vient, par exemple, d’un nombre plus grand de directives de routage pour la communication entre les processus et une latence variable dans le temps. La passage à l’échelle signifie que des ressources peuvent être ajoutées indéfiniment simplement en les reliant à l’infrastructure existante. Ces caractéristiques influencent directement la façon dont la performance des applications parallèles doit être analysée. Les techniques de visualisation traditionnelles pour cette analyse sont généralement basées sur des diagrammes de Gantt que disposent la liste des composants de l’application verticalement et metent la ligne du temps sur l’axe horizontal. Ces représentations visuelles ne sont généralement pas adaptés à l’analyse des applications exécutées en parallèle dans les grilles. La première raison est qu’elles n’ont pas été conçues pour offrir aux développeurs une analyse qui montre aussi la topologie du réseau des ressources. La deuxième raison est que les techniques de visualisation traditionnels ne s’adaptent pas bien quand des milliers d’entités doivent être analysés ensemble. Cette thèse tente de résoudre les problèmes des techniques traditionnelles dans la visualisation des applications parallèles. L’idée principale est d’exploiter le domaine de la visualisation de l’information et essayer d’appliquer ses concepts dans le cadre de l’analyse des programmes parallèles. Portant de cette idée, la thèse propose deux modèles de visualisation : les trois dimensions et le modèle d’agrégation visuelle. Le premier peut être utilisé pour analyser les programmes parallèles en tenant compte de la topologie du réseau. L’affichage lui-même se compose de trois dimensions, où deux sont utilisés pour indiquer la topologie et la troisième est utilisée pour représenter le temps. Le second modèle peut être utilisé pour analyser des applications parallèles comportant un très grand nombre de processsus. Ce deuxième modèle exploite une organisation hiérarchique des données utilisée par une technique appelée Treemap pour représenter visuellement la hiérarchie. Les implications de cette thèse sont directement liées à l’analyse et la compréhension des applications parallèles exécutés dans les systèmes distribués. Elle améliore la compréhension des modes de communication entre les processus et améliore la possibilité d’assortir les motifs avec cette topologie de réseau réel sur des grilles. Bien que nous utilisons abondamment l’exemple de la topologie du réseau, l’approche pourrait être adapté, avec presque pas de changements, à l’interconnexion fourni par un middleware d’une interconnexion logique. Avec la technique d’agrégation, les développeurs sont en mesure de rechercher des patterns et d’observer le comportement des applications à grande échelle. / Sistemas distribuídos tais como grids são usados hoje para a execução de aplicações paralelas com um grande número de processos. Algumas características desses sistemas são a presença de uma complexa rede de interconexão e a escalabilidade de recursos. A complexidade de rede vem, por exemplo, de largura de banda e latências variáveis ao longo do tempo. Escalabilidade é a característica pela qual novos recursos podem ser adicionados em um grid apenas através da conexão em uma infraestrutura pré-existente. Estas características influenciam a forma como o desempenho de aplicações paralelas deve ser analisado. Esquemas tradicionais de visualização de desempenho são usualmente baseados em gráficos Gantt com uma dimensão para listar entidades monitoradas e outra para o tempo. Visualizações como essa não são apropriadas para a análise de aplicações paralelas executadas em grid. A primeira razão para tal é que elas não foram concebidas para oferecer ao desenvolvedor uma análise que mostra a topologia dos recursos e a relação disso com a aplicação. A segunda razão é que técnicas tradicionais não são escaláveis quando milhares de entidades monitoradas devem ser analisadas conjuntamente. Esta tese tenta resolver estes problemas encontrados em técnicas de visualização tradicionais para a análise de aplicações paralelas. A idéia principal consiste em explorar técnicas da área de visualização da informação e aplicá-las no contexto de análise de programas paralelos. Levando em conta isto, esta tese propõe dois modelos de visualização: o de três dimensões e o modelo de agregação visual. O primeiro pode ser utilizado para analisar aplicações levando-se em conta a topologia da rede dos recursos. A visualização em si é composta por três dimensões, onde duas são usadas para mostrar a topologia e a terceira é usada para representar o tempo. O segundo modelo pode ser usado para analisar aplicações paralelas com uma grande quantidade de processos. Ela explora uma organização hierárquica dos dados de monitoramento e uma técnica de visualização chamada Treemap para representar visualmente a hierarquia. Os dois modelos representam uma nova forma de analisar aplicação paralelas visualmente, uma vez que eles foram concebidos para larga-escala e sistemas distribuídos complexos, como grids. As implicações desta tese estão diretamente relacionadas à análise e ao entendimento do comportamento de aplicações paralelas executadas em sistemas distribuídos. Um dos modelos de visualização apresentados aumenta a compreensão dos padrões de comunicação entre processos e oferece a possibilidade de observar tal padrão com a topologia de rede. Embora a topologia de rede seja usada, a abordagem pode ser adaptada sem grandes mudanças para levar em conta interconexões lógicas de bibliotecas de comunicação. Com a técnica de agregação apresentada nesta tese, os desenvolvedores são capazes de observar padrões de aplicações paralelas de larga escala. / Highly distributed systems such as grids are used today for the execution of large-scale parallel applications. Some characteristics of these systems are the complex resource interconnection that might be present and the scalability. The interconnection complexity comes from the different number of hops to provide communication among applications processes and differences in network latencies and bandwidth. The scalability means that the resources can be added indefinitely just by connecting them to the existing infrastructure. These characteristics influence directly the way parallel applications performance must be analyzed. Current traditional visualization schemes to this analysis are usually based on Gantt charts with one dimension to list the monitored entities and the other dimension dedicated to time. These visualizations are generally not suited to parallel applications executed in grids. The first reason is that they were not built to offer to the developer an analysis that also shows the network topology of the resources. The second reason is that traditional visualization techniques do not scale well when thousands of monitored entities must be analyzed together. This thesis tries to overcome the issues encountered on traditional visualization techniques for parallel applications. The main idea behind our efforts is to explore techniques from the information visualization research area and to apply them in the context of parallel applications analysis. Based on this main idea, the thesis proposes two visualization models: the three-dimensional and the visual aggregation model. The former might be used to analyze parallel applications taking into account the network topology of the resources. The visualization itself is composed of three dimensions, where two of them are used to render the topology and the third is used to represent time. The later model can be used to analyze parallel applications composed of several thousands of processes. It uses hierarchical organization of monitoring data and an information visualization technique called Treemap to represent that hierarchy. Both models represent a novel way to visualize the behavior of parallel applications, since they are conceived considering large-scale and complex distributed systems, such as grids. The implications of this thesis are directly related to the analysis and understanding of parallel applications executed in distributed systems. It enhances the comprehension of patterns in communication among processes and improves the possibility of matching this patterns with real network topology of grids. Although we extensively use the network topology example, the approach could be adapted with almost no changes to the interconnection provided by a middleware of a logical interconnection. With the scalable visualization technique, developers are able to look for patterns and observe the behavior of large-scale applications.

Bien que la consommation énergétique des processeurs a considérablement diminué, la demande pour des techniques visant à la réduire n’a jamais été aussi forte. En effet, la consommation énergétique des machines haute performance a crûproportionnellement à leurs accroissements en taille. Elle a atteint un tel niveau qu’elle doit être minimisée par tous les moyens. Les processeurs actuels peuvent changer au vol leurs fréquences d’exécution. Utiliser une fréquence plus faible peut mener à une réduction de leurs consommations énergétiques. Cette thèse recherche jusqu’à quel point cette fonctionnalité, appelé DVFS, peut favoriser cette réduction. Dans un premier temps, une analyse d’une machine simple est effectuée pour une meilleure compréhension des différents éléments consommateurs afin de focaliser les optimisations sur ces derniers. La consommation d’un processeur dépend de l’application qui est exécutée. Une analyse des applications est donc effectuée pour mieux comprendre leurs impacts sur cette dernière. Basés sur cette étude, plusieurs outils visant à réduire cette consommation ont été créés. REST, adapte la fréquence d’exécution au regard du comportement de l’application. Le second, UtoPeak, calcule la réduction maximum que l’on peut attendre grâce au DVFS. Le dernier, FoREST, est créé pour corriger les défauts de REST et obtenir cette réduction maximum de la consommationénergétique. Enfin, les applications scientifiques actuelles utilisent généralement plus d’unprocesseur pour leurs exécutions. Cette thèse présente aussi une première tentative de découverte de la borne inférieure sur la consommation énergétique dans ce nouvel environnement d’exécution / Over the past decade, processors have drastically reduced their power consumption. With each new processor generation, new features enhancing the processor energy efficiency are added. However, the demand for energy reductiontechniques has never been so high. Indeed, with the increasing size of high performance machines, their power and energy consumptions have grown accordingly. They have reached a point where they have to be reduced by all possible means.Current processors allow an interesting feature, they can change their operating frequency at run-time. As granted by transistor physics, lower frequency means lower power consumption and hopefully, lower energy consumption. This thesisinvestigates to which extent this processor feature, called DVFS, can be used to save energy. First, a simple machine is analyzed to have a complete understanding of the different power consumers and where optimizations can be focused. It will be demonstrated that only fans and processors allow run-time energy optimizations. Betweenthe two, the processor shows the highest consumption, therefore potentially exposing the higher potential for energy savings. Second, the power consumption of a processor depends on the applications being executed. However, there are as many applications as problems to solve. The focus is then put on applications to understand their impacts on energy consumption. Based on the gathered insights, multiple tools targeting energy savings on a single processor are created. REST, the most naive, tries to adapt the processor state to the stress generated by the application, hoping for energy reduction. The second, UtoPeak, computes the maximum energy reduction one can expect for any tool usingDVFS. It allows to evaluate the efficiency of such systems. The last one, FoREST, was created in order to correct all the flaws of REST and target maximum energy reduction. Last, scientific applications generally need more than one processor to be executed in a decent time. The thesis also presents a first attempt to compute a lower bound in energy reduction when considering this new execution context

[pt] Neste trabalho propomos usar o ALua, um mecanismo de comunicação orientado a eventos, baseado na linguagem interpretada Lua, para coordenação e desenvolvimento de aplicações paralelas distribuídas. ALua é um modelo de programação dual para aplicações paralelas distribuídas, que age como elemento de ligação, permitindo que partes pré- compiladas do programa sejam executadas em diferentes máquinas. Novas tecnologias em programação paralela, como computação em grade, e o interesse atual em computação distribuída para redes geográficas demandam novos níveis de flexibilidade, como o uso de estratégias de adaptação e a habilidade para um usuário interferir em uma computação sem a necessidade de interrompê-la. Além disso, devido a sua natureza assíncrona, a programação dirigida a eventos oferece um modelo apropriado para ambiente sujeitos a falhas e retardos, que são freqüentes no contexto de redes geográficas. Neste trabalho, mostramos que o ALua pode trazer a flexibilidade desejada, através de mecanismos de adaptação e monitoramento não só de aplicações, mas do próprio ambiente de execução, e ainda tirar proveito de sua natureza interpretada para permitir a intervenção do usuário na aplicação mesmo durante a sua execução. / [en] In this work we propose the use of Alua, an event-driven communication mechanism for coordinating and developing distributed parallel applications, based on the interpreted language Lua. Alua adopts a multilinguage programming model for distributed parallel applications, acting as a gluing element among precompiled program parts running on different machines. New developments in parallel programming, such as Grid computing, and current interest in wide-area distributed computing demand new levels of flexibility, such as the use of adaptive strategies and the ability for an user to interfer with a computation without having to stop it. Furthermore, because of its asynchronous nature, event-driven programming provides a suitable model for environments subject to failures and delays that are frequent in the context of geographically distributed computing. In this work we show that ALua can achieve the required flexibility through mechanisms for monitoring and adapting not only applications, but also the execution environment, and also exploit its interpretive nature to allow the programmer to modify the behavior of the application during its execution.