Optimizing Performance and Energy Efficiency in Massively Parallel Systems