DescriptionOur daily lives depend on data centers as they host popular Internet services and critical business applications. Managing these large and complex data centers is a challenging endeavor. We have designed and implemented three systems that simplify the management of data centers. The unifying characteristic of these systems is that they all rely on a limited number of experiments with real servers and workloads. The first system, called JustRunIt, replaces analytical modeling in assessing the performance, availability, and/or energy implications of potential management decisions or system configurations. The second system, called ACI, efficiently optimizes configurations as services evolve, by detecting and leveraging dependencies between configuration parameters. The last system, called MassConf, automatically configures server software for new users by leveraging configuration information from the existing users of the software. The evaluation shows that our systems significantly reduce the resources and time required to accomplish many management tasks. Given our experience and positive results with these three systems, we conclude that experiment-based management has the potential to be very useful in practice.