Online busy time scheduling with flexible jobs

2024-05-14

arXiv.org

AI Summary

This paper studies the online busy-time scheduling problem in cloud computing: scheduling jobs with heterogeneous release times, deadlines, and processing requirements on identical parallel machines to minimize total machine activation time (i.e., energy-aware scheduling). We propose novel online algorithms—integrating greedy selection with piecewise scheduling—for both unbounded and bounded numbers of processors (g). Our theoretical contributions are threefold: (1) For unbounded (g), we tighten the competitive ratio from (O(log g)) to (Theta(1)), the first constant-factor guarantee; (2) For bounded (g) without prior knowledge of job parameters, we achieve the first constant competitive ratio, significantly improving the best-known upper bound; (3) We extend all results to the agreeable job class (where release times and deadlines are similarly ordered). All algorithms are rigorously analyzed via competitive analysis, yielding stronger theoretical guarantees than prior work.

Technology Category

• Planning, Routing, and Scheduling: Scheduling under Uncertainty
• Machine Learning: Online Learning & Bandits
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization

Application Category

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

Abstract

We present several competitive ratios for the online busy time scheduling problem with flexible jobs. The busy time scheduling problem is a fundamental scheduling problem motivated by energy efficiency with the goal of minimizing the total time that machines with multiple processors are enabled. In the busy time scheduling problem, an unbounded number of machines is given, where each machine has $g$ processors. No more than $g$ jobs can be scheduled simultaneously on each machine. A machine consumes energy whenever at least one job is scheduled at any time on the machine. Scheduling a single job at some time $t$ consumes the same amount of energy as scheduling $g$ jobs at time $t$. In the online setting, jobs are revealed when they are released. We consider the cases where $g$ is unbounded and bounded. In this paper, we revisit the bounds of the unbounded general setting from the literature and tighten it significantly. We also consider agreeable jobs. For the bounded setting, we show a tightened upper bound. Furthermore, we show the first constant competitive ratio in the bounded setting that does not require lookahead.

Problem

• Minimizing machine busy times in cloud computing systems
• Scheduling flexible jobs with release times and deadlines
• Developing online algorithms with tight competitive ratios

Innovation

• Online scheduling algorithm for cloud energy optimization
• Tight competitive ratio analysis for multiple variants
• Lookahead improvement for arbitrary job scheduling