The present invention discloses an energy-saving deployment method of a virtual CDN. According to the historical flow data of the virtual CDN and the prediction model (ARIMA) in the controller, the network peak flow in the next time period is predicted. Next, the scale of the virtual CDN system at the next moment is calculated according to the peak flow. Meanwhile, several redundant servers are added to correct the prediction error. The network flow is aggregated to the desired virtual servers based on the calculation of the controller through a load balancer. In this way, the utilization rate of the virtual CDN system can be increased, and the energy consumed due to the higher utilization rate of the CDN system is saved.