Most components of High Performance Computing systems, either hardware or software, come with many tunable parameters and their parametrization can have a significant impact on their performance. For optimal performance, the most adapted parametrization for each application running on the cluster must be determined and used. However, this parametrization is difficult to find because of the complexity of the relationship between each component and the lack of insight on the system’s behavior.In this thesis, we remove the complex task of tuning the system manually or through theoretical models, by exploring auto-tuning methods relying on black-box optimization to find the system’s optimal parameters. We provide a comprehensive comparison of t...