Learning Augmented Energy Minimization via Speed Scaling
This addresses energy efficiency in data centers, an incremental advance by integrating ML predictions into a classic optimization problem.
The paper tackles the online speed scaling problem for data center energy minimization by incorporating machine learning predictions about future workload, proposing an algorithm that outperforms traditional online algorithms when predictions are accurate while maintaining provable guarantees when predictions are poor.
As power management has become a primary concern in modern data centers, computing resources are being scaled dynamically to minimize energy consumption. We initiate the study of a variant of the classic online speed scaling problem, in which machine learning predictions about the future can be integrated naturally. Inspired by recent work on learning-augmented online algorithms, we propose an algorithm which incorporates predictions in a black-box manner and outperforms any online algorithm if the accuracy is high, yet maintains provable guarantees if the prediction is very inaccurate. We provide both theoretical and experimental evidence to support our claims.