I. Introduction
IT IS COMMON practice to tune the performance of a system through the optimization of some function or measure that is not directly related with the real objective of that system. This is the case of high-performance computing (HPC) systems, which are usually configured to try to minimize the makespan, waiting times, or some other easily measurable criteria. However, the real objective of an HPC center is not to minimize the makespan or similar criteria but to maximize the user's satisfaction regarding the system behavior and, at the same time, minimize cost through the optimal use of available resources.