Systems based on the MapReduce programming model are emerging as a central tool for deploying jobs that process large datasets in parallel. However the configuration of MapReduce systems is a complex process and at the moments it's left up to the user. These ad-hoc configuration methods make it difficult for small companies to take advantage of the growth of cloud computing solutions that provide resources as a service. Furthermore, the definition of SLAs becomes a complicated process for the user and the service provider as well. We propose a control theoretical approach to solving these problems. This implies the development of a general model that captures the dynamics of MapReduce systems. Finally, we intend to provide novel control methods that ease the configuration process and guarantee service level objectives such as constraints on system performance (execution times) and dependability (latency, availability) while optimizing resource consumption.