Abstract : Clusters are massively used through Resource Management Systems with a static allocation of resources for a bounded amount of time. Such an approach leads to a coarse-grain exploitation of the architecture and an increase of the job completion times since most of the scheduling policies rely on users estimates and do no consider the real needs of applications in terms of both resources and times. Encapsulating jobs into VMs enables to implement finer scheduling policies through cluster-wide context switches: a permutation between VMs present in the cluster. It results a more flexible use of cluster resources and relieve end-users of the burden of dealing with time estimates. Leveraging the Entropy framework, this paper introduces a new infrastructure enabling cluster-wide context switches of virtualized jobs to improve resource management. As an example, we propose a scheduling policy to execute a maximum number of jobs simultaneously, and uses VM operations such as migrations, suspends and resumes to resolve underused and overloaded situations. We show through experiments that such an approach improves resource usage and reduces the overall duration of jobs. Moreover, as the cost of each action and the dependencies between them is considered, Entropy reduces, the duration of each cluster-wide context switch by performing a minimum number of actions, in the most efficient way.