Energy Optimization by Fan Speed Control for Data Centers Open Access
Downloadable ContentDownload PDF
Currently, energy efﬁciency for data centers is an attractive problem. For data centers, good scheduling algorithms must not only satisfy the Quality of Service (QoS) and but also try to minimize the total energy. However, such algorithmsdonot always consider the effect of the cooling system,which consumes a signiﬁcant amount of energy by itself. An algorithm that can balance the QoS and energy cost of the CPU and network components might cause an unexpected energy increase in the cooling system. According to , the energy of cooling system in data centers can be as large as 50 percent of the overall consumption. This thesis investigates algorithms that can balance the overall energy consumption with QoS by considering a cluster of compute servers, each of which is cooled by fans. The impact of CPU power on the cooling system can be predicted by a thermal model, and by relating to the power model of fan, we get the relation between CPU power and the fan power. Workload Adaptive Energy-Latency Optimization in Server Farms using Server Low-Power States (WASP) is a previously proposed algorithm that can adjust itself dynamically according to workload. This thesis exmplores the impact of WASP on the cooling system under different sizes of jobs and utilization levels, and proposes a pre-cooling based optimization for the fan speed to improve WASP.