Competitive analysis of preemptive single-machine scheduling
Introduction
Many scheduling problems with the objective of minimizing the average (weighted) completion time can be approximated efficiently to within any constant arbitrarily close to 1, i.e. they have polynomial time approximation schemes (PTAS’s) [1]. However, in the online setting where we assume jobs arrive over time and the schedule is constructed without knowledge of jobs arriving in the future, the best known approximation ratio is often much worse than what we get from offline computing. For some problems, lower bounds show that no algorithm that works online can have a small approximation ratio, but for other problems these bounds are missing and they may have simple, efficient online algorithms with ratios close to 1. In this paper, we study the single-machine scheduling problem with preemption of jobs and the average weighted completion time objective. This problem does have a PTAS and the best known lower bound on the competitive ratio (online approximation ratio) is only 1.073 [5]. We give a simple, -time online algorithm and prove that its competitive ratio is at most 1.57. (A slightly different proof is given in the author’s Ph.D. Thesis [11].) Our proof is based on the elegant technique of modified instances.
An instance of our problem is given by a set and for each a positive processing time , non-negative weight and non-negative release time . We assume all numbers are rational and write . A schedule determines for each moment in time which job is processed on the machine. Each job needs to be processed for exactly time units and must not start before time . Preemption is allowed, i.e., we may split the units over several intervals. Given a feasible schedule , we denote the completion time of job by and denote the weighted sum of completion times by . We denote the minimum weighted sum of completion times over all feasible schedules for by .
In the non-preemptive setting no deterministic online algorithm can be better than 2-competitive [7]. A 2-competitive algorithm was given by Anderson and Potts [3]. A simplified proof by Queyranne [9] shows that the ratio holds with respect to the preemptive problem as well. Goemans, Wein and Williamson (cited as a personal communication in [10]) noted that scheduling at any point in time the job with highest ratio of weight over processing time gives a 2-competitive algorithm. This is sometimes called the preemptive weighted shortest processing time rule (WSPT). Megow [8] considers the shortest weighted remaining processing time rule (SWRPT) in which at any moment the job with largest ratio of weight over remaining processing time is processed. She shows that the ratio of this algorithm is at most 2 but not better than 1.21 and conjectures that it is much smaller than 2. In contrast to the non-preemptive problem, the best lower bound on the ratio for the preemptive problem is only 1.073 [5]. A randomized online algorithm with ratio 4/3 was given by Schulz and Skutella [10]. For an overview of approximation ratios for average completion time scheduling see [2], [4], [12].
The mean busy time of a job in a schedule is the average time at which it is processed. Formally, let be the indicator function of job in a given schedule, i.e., if job is processed at time and otherwise. Then, the mean busy time of job is defined by Observe that for any job . Equality holds if and only if job is not preempted. Given an instance , we say that functions for all define a relaxed schedule if
- (i)
for all , and
- (ii)
for all .
Section snippets
Algorithm and proof
Our algorithm has a parameter and it applies the preemptive WSPT rule with the restriction that a job is not preempted at a time if it can be completed before time . It matches with the preemptive WSPT rule for .
: At any moment that a job completes, start the job with largest ratio of among the available jobs. At any moment that a new job is released and the remaining processing time of the currently running job is more than , start the job with largest ratio
References (12)
- et al.
Designing PTASs for min-sum scheduling problems
Discrete Applied Mathematics
(2006) - F. Afrati, E. Bampis, C. Chekuri, D. Karger, C. Kenyon, S. Khanna, I. Milis, M. Queyranne, M. Skutella, C. Stein, M....
- et al.
Online scheduling of a single machine to minimize total weighted completion time
Mathematics of Operations Research
(2004) - et al.
LP-based online scheduling: from single to parallel machines
Mathematical Programming
(2009) - et al.
Lower bounds for on-line single-machine scheduling
- M.X. Goemans, Improved approximation algorithms for scheduling with release dates, in: Proc. 8th Symp. on Discrete...
Cited by (13)
On-line scheduling with monotone subsequence constraints
2019, Theoretical Computer ScienceCompetitive Kill-and-Restart and Preemptive Strategies for Non-clairvoyant Scheduling
2023, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)On competitive analysis for polling systems
2020, Naval Research LogisticsModeling single machine preemptive scheduling problems for computational efficiency
2020, Annals of Operations Research