The energy dissipated as heat for each utilization level of a data center server is empirically measured and stored as the thermal-profile. These thermal-profiles are used to predict the outlet temperatures of the related servers for current and future utilization. The predicted outlet temperature is an important parameter for energy efficient thermal-aware workload scheduling and workload migration in green data centers. This paper presents three models for outlet temperature prediction on virtualized data center servers based on thermal-profile. The best case scenario managed to predict the outlet temperature with a negligible error of 0.3 degree Celsius.Keywords: data center; servers; monitoring; virtualization; thermal-profile; thermal-prediction;
INTRODUCTIONMonitoring systems in data centers look-over the environment and performance of tens and hundreds of servers periodically. There are various parameters (e.g., temperature and utilization) monitored for each server. This data is required by data center infrastructure management (DCIM) [1] tools and workload scheduling systems [2] to achieve energy efficient and proficient utilization on data center servers. Apart from the idle energy consumption, the total energy consumed and dissipated as heat by each server is increased for every utilization level increment and vice versa. The energy usage of a virtualized server involves the virtualized instances of operating systems called virtual machines (VMs). From now on, the word server is used interchangeably with virtualized server.As long as the hardware configuration of a server is not altered, each heterogeneous server will dissipate a different but certain amount of heat for each discrete utilization level. This symbolic heat is empirically measured and stored as a thermal-profile for each server. The thermal-profile is used to predict the outlet temperature of a server by a given value of CPU usage through thermal-prediction modeling. A thermal prediction model eliminates at least one of the parameters e.g., outlet temperature, from the data center monitoring system without the loss of performance and accuracy. Similarly, the number of thermal sensors used for outlet temperature monitoring is reduced. The ability of a monitoring system to generate accurate thermal predictions offline, makes the prediction to become an essential parameter for thermal-aware workload scheduling and thermal-aware workload migration for load balancing.Based upon thermal-profile, this paper presents multiple prediction models to predict outlet temperature of data center servers along with a matrix comparison among these models. The test results show that the outlet