TY - GEN

T1 - Real-Time k-bounded preemptive schedulingc

AU - Albagli-Kim, Sivan

AU - Schiebcr, Baruch

AU - Shachnai, Hadas

AU - Tamir, Tami

PY - 2016/1/1

Y1 - 2016/1/1

N2 - We consider a variant of the classic real-time scheduling problem, which has natural applications in cloud computing. The input consists of a set of jobs, and an integer parameter k > 1. Each job is associated with a processing time, a release time, a due-date and a positive weight. The goal is to feasibly schedule a subset of the jobs of maximum total weight on a single machine, such that each of the jobs is preempted at most k times. Our theoretical results for the real-time k-bounded preemptive scheduling problem include hardness proofs, as well as algorithms for subclasses of instances, for which we derive constant-ratio performance guarantees. We bridge the gap between theory and practice through a comprehensive experimental study, in which we also test the performance of several heuristics for general instances on multiple parallel machines. We use in the experiments a linear programming relaxation to upper bound the optimal solution for a given instance. Our results show that while fc-bounded preemptive scheduling is hard to solve already on highly restricted instances, simple priority-based heuristics yield almost optimal schedules for realistic inputs and arbitrary values of k.

AB - We consider a variant of the classic real-time scheduling problem, which has natural applications in cloud computing. The input consists of a set of jobs, and an integer parameter k > 1. Each job is associated with a processing time, a release time, a due-date and a positive weight. The goal is to feasibly schedule a subset of the jobs of maximum total weight on a single machine, such that each of the jobs is preempted at most k times. Our theoretical results for the real-time k-bounded preemptive scheduling problem include hardness proofs, as well as algorithms for subclasses of instances, for which we derive constant-ratio performance guarantees. We bridge the gap between theory and practice through a comprehensive experimental study, in which we also test the performance of several heuristics for general instances on multiple parallel machines. We use in the experiments a linear programming relaxation to upper bound the optimal solution for a given instance. Our results show that while fc-bounded preemptive scheduling is hard to solve already on highly restricted instances, simple priority-based heuristics yield almost optimal schedules for realistic inputs and arbitrary values of k.

UR - http://www.scopus.com/inward/record.url?scp=84964047821&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:84964047821

T3 - Proceedings of the Workshop on Algorithm Engineering and Experiments

SP - 127

EP - 137

BT - 18th Workshop on Algorithm Engineering and Experiments 2016, ALENEX 2016

A2 - Goodrich, Michael

A2 - Mitzenmacher, Michael

PB - Society for Industrial and Applied Mathematics Publications

T2 - 18th Workshop on Algorithm Engineering and Experiments 2016, ALENEX 2016

Y2 - 10 January 2016

ER -