Distributed cyber-physical systems(CPS) deployed in performance-critical applications are imposed stringent constraints such as the real-time requirement and energy consumption budget. However, the performance of CPS is inevitably undermined by various physical uncertainties, for example, the stochastic environment noise and dynamics of the physical phenomenon. Both of those uncertainties might lead to task overload and the power overconsumption as task execution times are unpredictable. While recent feedback control scheduling have shown promise to meet the end-to-end deadlines by adaptively adjusting the task loading rate, little work has focused on both the system service rate and the CPU execution modes as a whole to minimize the system power consumption. This paper presents a feedback control based algorithm that adaptively maintains desired task service rate for real-time requirement while minimizes the power consumption by assigning proper execution modes on processor for each node. The theoretic analysis and numerous simulations demonstrate that the proposed scheme can provide robust real-time guarantees and reduce the power consumption when the task workload vary significantly at run-time.