In the following, we consider the problem of minimizing the energy consumption needed for executing a set of real-time tasks scheduled on a fixed number of identical processors. The scheduling is preemptive and follows the global EDF policy. ``Global'' scheduling algorithms, on the contrary to partitioned algorithms, allows different instances of the same task (also called jobs or processes) to be executed upon different processors. Each process can start its execution on any processor and may migrate at run-time from one processor to another if it gets preempted by smaller-deadline processes. We first tackle the problem of choosing the smallest admissible processor frequency for the set of CPUs such that all deadlines will be met considering the worst-case workload. The procedure is performed off-line and provides a static result in the sense that the computed speed does not change over time. Such a static solution is necessary, however, due to the discrepancy between worst-Case Execution Times (WCET) and Actual-Case Execution Times (ACET), it usually leads to very conservative results. In a second step, we thus propose an on-line ``slack reclaiming'' scheme that monitors task executions and take advantage of unused CPU time to further reduce frequency.