In this work, a model of computation for shared memory parallelism is presented. To address fundamental constraints of modern memory systems, the presented model constrains how parallelism interacts with memory access patterns and in doing so provides a method for design and analysis of algorithms that estimates reliable execution time based on a few architectural parameters. This model is presented as an alternative to modern thread based models that focus on computational concurrency but rely on reactive hardware policies to hide and amortize memory latency. Since modern processors use reactive mechanisms and heuristics to deduce the data access requirement of computations, the memory access costs of these threaded programs may be difficult to predict reliably. This research presents the Queue Streaming Model (QSM) that aims to address these shortcomings by providing a prescriptive mechanism to achieve latency-amortized and predictable-cost data access. Further, the work presents application of the QSM to algorithms commonly used in a number of applications. These algorithms include structured regular computations represented by merge sort, unstructured irregular computations represented by sparse matrix dense vector multiplication, and dynamic computations represented by MapReduce. The analysis of these algorithms reveal architectural tradeoffs between memory system bottlenecks and algorithm design. The techniques described in this dissertation reveal a general software approach that could be used to construct more general irregular applications, provided they can be transformed into a relational query form. It demonstrates that the QSM can be used to design algorithms that enhance utilization of memory system resources by structuring concurrency and memory accesses such that system bandwidths are balanced and latency is amortized. Finally, the benefit of applying the QSM algorithm to the Euler inviscid flow solver is demonstrated through experiments on the Intel(R) Xeon(R) E5-2680 v2 processor using ten cores. The transformation produced a speed-up of 25% over an optimized OpenMP implementation having identical computational structure.

A scale-linking, experimental study complemented by an analytical model was carried out to investigate the influence of fiber orientation on the crack-opening behavior of strain-hardening cement-based composites (SHCC). Three SHCC compositions were investigated with polyvinyl alcohol (PVA) and ultra-high molecular weight polyethylene (UHMWPE) fibers in combination with normal- and high-strength matrices. The micromechanical experiments with fiber inclinations of 0◦, 30◦, 45◦, and 60◦ involved fiber embedment in plain and fiber-reinforced specimens. The experimentally derived micromechanical parameters were input into an analytical crack-bridging model to assess the upscaling accuracy of the micromechanical results by comparing the predicted crack-bridging laws to the single-crack opening behavior of equivalent miniature SHCC specimens with controlled fiber orientation. This study yields new insights into the effect of fiber orientation on the crackbridging properties of different types of SHCC, assesses the link between micromechanical and composite scale properties, offers a solid experimental basis for refining the analytical models, and developing anisotropic materials models for SHCC in dependence on fiber orientation.

La programmation parallèle et les algorithmes data-parallèles sont depuis plusieurs décennies les principales techniques de soutien l'informatique haute performance. Comme toutes les propriétés non-fonctionnelles du logiciel, la conversion des ressources informatiques dans des performances évolutives et prévisibles implique un équilibre délicat entre abstraction et automatisation avec une précision sémantique. Au cours de la dernière décennie, de plus en plus de professions ont besoin d'une puissance de calcul très élevée, mais la migration des programmes existants vers une nouvelle configuration matérielle et le développement de nouveaux algorithmes à finalité spécifique dans un environnement parallèle n'est jamais un travail facile, ni pour les développeurs de logiciel, ni pour les spécialistes du domaine. Dans cette thèse, nous décrivons le travail qui vise à simplifier le développement de programmes parallèles, en améliorant également la portabilité du code de programmes parallèles et la précision de la prédiction de performance d'algorithmes parallèles pour des environnements hétérogènes. Avec ces objectifs à l'esprit, nous avons proposé un modèle de transition nommé SGL pour la modélisation des architectures parallèles hétérogènes et des algorithmes parallèles, et une mise en œuvre de squelettes parallèles basés sur le modèle SGL pour le calcul haute performance. SGL simplifie la programmation parallèle à la fois pour les machines parallèles classiques et pour les nouvelles machines hiérarchiques. Il généralise les primitives de la programmation BSML. SGL pourra plus tard en utilisant des techniques de Model-Driven pour la génération de code automatique á partir d'une fiche technique sans codage complexe, par exemple pour le traitement de Big-Data sur un système hétérogène massivement parallèle. Le modèle de coût de SGL améliore la clarté de l'analyse de performance des algorithmes, permet d'évaluer la performance d'une machine et la qualité d'un algorithme / Parallel programming and data-parallel algorithms have been the main techniques supporting high-performance computing for many decades. Like all non-functional properties of software, the conversion of computing resources into scalable and predictable performance involves a delicate balance of abstraction and automation with semantic precision. During the last decade, more and more professions require a very high computing power. However, migrating programs to new hardware configuration or developing new specific-purpose algorithms on a parallel environment is never an easy work, neither for software developers nor for domain specialists. In this thesis we describe work that attempts to improve the simplicity of parallel program development, the portability of parallel program code, and the precision of parallel algorithm performance prediction for heterogeneous environments. With these goals in mind we proposed a bridging model named SGL for modelling heterogeneous parallel architectures and parallel algorithms, and an implementation of parallel skeletons based on SGL model for high-performance computing. SGL simplifies the parallel programming either on the classical parallel machines or on the novel hierarchical machines. It generalizes the BSML programming primitives. SGL can be served later with model-driven techniques for automatic code generation from specification sheet without any complex coding, for example processing Big Data on the heterogeneous massive parallel systems. The SGL cost model improves the clarity of algorithms performance analysis; it allows benchmarking machine performance and algorithm quality