@online{Antoniadis2112.03082,
TITLE = {A Novel Prediction Setup for Online Speed-Scaling},
AUTHOR = {Antoniadis, Antonios and Jabbarzade Ganje, Peyman and Shahkarami, Golnoosh},
LANGUAGE = {eng},
URL = {https://arxiv.org/abs/2112.03082},
EPRINT = {2112.03082},
EPRINTTYPE = {arXiv},
YEAR = {2022},
ABSTRACT = {Given the rapid rise in energy demand by data centers and computing systems<br>in general, it is fundamental to incorporate energy considerations when<br>designing (scheduling) algorithms. Machine learning can be a useful approach in<br>practice by predicting the future load of the system based on, for example,<br>historical data. However, the effectiveness of such an approach highly depends<br>on the quality of the predictions and can be quite far from optimal when<br>predictions are sub-par. On the other hand, while providing a worst-case<br>guarantee, classical online algorithms can be pessimistic for large classes of<br>inputs arising in practice.<br> This paper, in the spirit of the new area of machine learning augmented<br>algorithms, attempts to obtain the best of both worlds for the classical,<br>deadline based, online speed-scaling problem: Based on the introduction of a<br>novel prediction setup, we develop algorithms that (i) obtain provably low<br>energy-consumption in the presence of adequate predictions, and (ii) are robust<br>against inadequate predictions, and (iii) are smooth, i.e., their performance<br>gradually degrades as the prediction error increases.<br>},
}