Enhanced precision data center server power consumption model with temperature estimation based on CPU operating statues

Abstract

To mitigate energy consumptions in data centers, the accurate establishment of a server power consumption model is imperative. Traditional server power consumption model, which rely solely on CPU utilization, often overlook the CPU temperature and operating statues inherent characteristics, resulting in substantial forecast errors. In response to this gap, a novel enhanced precision Power consumption Model based on Temperature estimation considering CPU working state (PMTC) model, is proposed based on the identification of CPU operating statuses. By incorporating temperature variables at the initial model construction phase, the PMTC model effectively captures the delayed dynamic characteristics of server power consumption changes that are influenced by the lagging adjustments in CPU core temperatures, thereby eliminating temperature-related modeling inaccuracies. In the subsequent power forecasting stage, the PMTC model accurately identifies specific CPU operating statues, which facilitate precise estimations of the CPU core temperature, thus circumventing the implementation challenges associated with additional measurements of temperature variables. To validate the efficacy of the proposed PMTC model against traditional server power consumption models, a dedicated server power consumption testbed was established. The results demonstrate that the PMTC model, by incorporating the temperature-related delayed dynamic characteristics of server power consumption change without augmenting the dimensions of input data, significantly reduces modeling calculation errors.