Energy-Efficient Scheduling with Predictions

TL;DR

The paper tackles energy-efficient scheduling with predictions by introducing the General Energy-Efficient Scheduling with Predictions (GESP) model and a two-phase algorithm (TPE) that combines offline planning on predicted jobs with online handling of actual arrivals. By proving that the algorithm achieves $1+\varepsilon$-consistency while remaining $O(1)$-robust, and that the competitive ratio c(η) degrades smoothly with prediction error η, the work provides strong, tunable guarantees beyond prior deadlines-focused results. The framework applies to diverse GES objectives, including energy-plus-flow-time and deadlines, and extends to approximate predictions via a shift-tolerance extension (TPE-S) with rigorous smoothness-based analysis. Empirical results on real and synthetic datasets corroborate the theoretical findings, showing improved performance when predictions are accurate and bounded degradation under errors. Overall, the work offers a practical, flexible approach for leveraging predictions to improve energy-efficient scheduling in data-centers and similar systems.

Abstract

An important goal of modern scheduling systems is to efficiently manage power usage. In energy-efficient scheduling, the operating system controls the speed at which a machine is processing jobs with the dual objective of minimizing energy consumption and optimizing the quality of service cost of the resulting schedule. Since machine-learned predictions about future requests can often be learned from historical data, a recent line of work on learning-augmented algorithms aims to achieve improved performance guarantees by leveraging predictions. In particular, for energy-efficient scheduling, Bamas et. al. [BamasMRS20] and Antoniadis et. al. [antoniadis2021novel] designed algorithms with predictions for the energy minimization with deadlines problem and achieved an improved competitive ratio when the prediction error is small while also maintaining worst-case bounds even when the prediction error is arbitrarily large. In this paper, we consider a general setting for energy-efficient scheduling and provide a flexible learning-augmented algorithmic framework that takes as input an offline and an online algorithm for the desired energy-efficient scheduling problem. We show that, when the prediction error is small, this framework gives improved competitive ratios for many different energy-efficient scheduling problems, including energy minimization with deadlines, while also maintaining a bounded competitive ratio regardless of the prediction error. Finally, we empirically demonstrate that this framework achieves an improved performance on real and synthetic datasets.

Figures (3)

• Figure 1: The competitive ratio achieved by our algorithm, TPE-S, and the benchmark algorithm, as a function of the error parameter $\sigma$ (from left-most to the second from the right), and the competitive ratio of TPE-S for a larger range of $\sigma$, as a function of $\sigma$ (right-most).
• Figure 2: The competitive ratio achieved by our algorithm, TPE-S, as a function of the shift tolerance parameter $\eta^{\text{shift}}$ (left) and as a function of the confidence parameter $\lambda$ (right).
• Figure 3: The competitive ratio achieved by our algorithm, TPE-S and the benchmark algorithm as a function of the shift tolerance $\eta^{shift}$ (row 1) and as a function of the confidence parameter $\lambda$ (row 2).

Theorems & Definitions (46)

• Lemma 3.0
• Lemma 3.0
• proof
• Corollary 3.0
• Theorem 3.1
• Corollary 3.1
• Proposition A.1
• proof
• Lemma A.1
• Lemma A.2