Current students

Bayesian Optimization (BO) is a promising approach for high performance computing (HPC) cluster configuration due to its proved effectiveness at tackling the complex nature of software running on HPC, which depends on both model- and system-level configuration parameters. On the other hand, machine learning techniques can provide useful knowledge about the HPC structure at hand. The proposed research work aims at combining the best features of both methods to have accurate, fast prediction for the choice of such parameters.

Because of its strategic importance in solution of grand challenge problems and in the promotion of social and economic development, high performance computing (HPC) has been one of the most active research areas in both computer science and engineering over the last 40 years. HPC systems are complex and introduce also very complex software stacks that provide a large number of parameters influencing the final application performance. The aim of this thesis is to develop machine learning based models and a Bayesian optimization (BO) method to optimize the performance of a Drug Discovery application running in a HPC cluster.
The techniques used and developed in this research fall under performance evaluation theory and applications, machine learning and Bayesian optimization. The research will be based on machine learning approaches applied to predict the performance of drug discovery applications and Bayesian optimization methods. The optimization final goal is to identify the minimum cost configuration of a HPC cluster which will provide performance guarantees (i.e., a deadline) for running the drug discovery simulations.