Cooling system consumes more than 35% of total electricity in most data centers. The provided cooling normally exceeds the actual demand of IT equipment in order to assure the safe operation, resulting in a low energy efficiency. In this paper, a novel method based on demand response was proposed to precisely control the cooling supply, and the energy saving potential was assessed systematically. Compared to the reference case, in which the cooling demand is determined by assuming all of servers are in the running status, when the cooling demand was determined based on the measured dynamic IT load at room level, row level, rack level and server level, it can be reduced by 7.9%, 14.2%, 15.6% and 17.9% respectively for the random selected 48 h. In addition, IT load shifting also has a big potential to save energy, as it can make the cooling system working at a higher energy efficiency, which varies with loads. Two cases were studied: even distribution of IT load and optimized IT load shifting. Compared to the best case that determines the cooling demand according to the IT load at server level, they can further reduce the electricity consumption of cooling systems by 0.9%, and 1.2%. (C) 2017 Elsevier Ltd. All rights reserved.