Load Balancing with Duration Predictions

2025-12-13

AI Summary

This paper studies prediction-augmented dynamic load balancing: minimizing the maximum load or the $l_p$-norm of machine loads in heterogeneous environments with jobs arriving and departing over time. Recognizing that job processing times are only partially predictable and subject to prediction errors in practice, we propose the first theoretical framework that quantitatively links prediction error to competitive ratio. We design online load assignment algorithms that are robust to prediction inaccuracies, achieving a smooth degradation of competitive ratio with increasing error—avoiding the sharp performance drop typical of classical clairvoyant or completely oblivious models. We establish matching lower bounds, demonstrating optimality of our approach. Notably, even under polynomial-scale prediction errors, our algorithms retain $mathrm{polylog}$-competitive guarantees, thereby bridging the theoretical gap between fully predictive and entirely prediction-free settings.

Technology Category

• Machine Learning: Online Learning & Bandits
• Planning, Routing, and Scheduling: Learning for Planning and Scheduling
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs
• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

We study the classic fully dynamic load balancing problem on unrelated machines where jobs arrive and depart over time and the goal is minimizing the maximum load, or more generally the l_p-norm of the load vector. Previous work either studied the clairvoyant setting in which exact durations are known to the algorithm, or the unknown duration setting in which no information on the duration is given to the algorithm. For the clairvoyant setting algorithms with polylogarithmic competitive ratios were designed, while for the unknown duration setting strong lower bounds exist and only polynomial competitive factors are possible. We bridge this gap by studying a more realistic model in which some estimate/prediction of the duration is available to the algorithm. We observe that directly incorporating predictions into classical load balancing algorithms designed for the clairvoyant setting can lead to a notable decline in performance. We design better algorithms whose performance depends smoothly on the accuracy of the available prediction. We also prove lower bounds on the competitiveness of algorithms that use such inaccurate predictions.

Problem

• Bridging gap between clairvoyant and unknown job duration settings
• Designing algorithms using job duration predictions for load balancing
• Analyzing performance based on prediction accuracy in dynamic scheduling

Innovation

• Uses duration predictions for load balancing
• Smooth performance based on prediction accuracy
• Proves lower bounds for inaccurate prediction algorithms