In the context of the continuous growth of the performance of modern computing systems, the problem of controlling power becomes more urgent. Exceeding the set peak values of power can lead to serious failures in the computing system and the emergency termination of user jobs with the loss of the obtained calculation results. It is necessary to predict the possible excess of the power values set by the administrators of the computing system and take measures to prevent it. This is important because the built-in power control mechanisms are used to ensure the ‘‘survivability’’ of the system in cases where it is no longer possible to take measures to stabilize its overload. The main priority is to maintain operability, rather than achieve peak performance values. Solving the problem of maintaining the performance of a computing system in conditions of limited power, it is important to be able to predict and prevent possible threshold values from being exceeded. The accuracy of calculating the power depends on how thoroughly the characteristics of energy-consuming components are taken into account in the model. The energy-intensive components used in clusters often make a significant contribution to the peak power of the entire computing system. Therefore, it is necessary to consider their contribution to the total power of the computing system. The main purpose of this research is to select a model that can be used to create realistic forecasts of peak power based on data from the specifications of computer system elements. Authors have classified the existing models and selected the most suitable one for use. The considered model is a combination of existing models and allows combining the required accuracy of forecasting power and the ease of obtaining data for calculation from the specifications of computer system components.
Kiselev et al. (Sat,) studied this question.