Needs in computing power are increasing steadily. In order to meet those needs, new architectures have appeared. Computing grids and clusters are gathering the computing power of several machines, and multi-core processors are offering more computing power without the need to increase the frequency and electrical power usage. Nonetheless, The study of those new architectures is not an easy task. In order to study computing grids researchers have dedicated platforms. But the conduct of experiments on such platforms is a difficult process. In the case of multi-core processors, their behavior is not well known. In this thesis we put forward two tools dedicated to the study of new architectures. The first one, Expo, is an experiment management software for grid architectures. Expo allows for automated experiment conduct while still trying to guarantee its sound unrolling. Expo also provides a language to describe experiments on grid infrastructures. The second one, PaSTeL, is dedicated to the study of interactions between architectures and execution strategies. Highly tunable, it grants user with means to study in depth the behavior of a parallel solution running on a given architecture. Both tools have been validated during the development phase, but also during an experimental study of a work stealing engine on several multi-core architectures.
