Data-centers have recently experienced a fast growth in energy demand, mainly due to cloud computing, a paradigm that lets the users access shared computing resources (e.g., servers, storage, etc.). Several techniques have been proposed in order to alleviate this problem, and numerous power models have been adopted to predict the servers' power consumption. Some of them consider many server resources, some others account for only the CPU, that has proven to be the component responsible for the largest part of a server's power consumption. All these models work with generally inaccurate input parameters. However, none of them takes into account the effects of such inaccuracy on the model outputs. This paper investigates how epistemic (parametric) uncertainty affects a power model. Studying the impact of epistemic uncertainty on power consumption models makes it possible to consider loads with a probability density while investigating the battery depletion time or the amount of energy required for a given task.
Gribaudo, M., Pinciroli, R., & Trivedi, K. (2018). Epistemic Uncertainty Propagation in Power Models. Electronic Notes in Theoretical Computer Science, 337, 67–86. https://doi.org/10.1016/j.entcs.2018.03.034