An EPTAS for machine scheduling with bag-constraints

TL;DR

This paper presents the first Efficient Polynomial-Time Approximation Scheme (EPTAS) for machine scheduling with bag-constraints, improving previous PTAS results by relaxing constraints and enabling independent scheduling of large and small jobs.

Contribution

The paper introduces a novel EPTAS for machine scheduling with bag-constraints, utilizing new insights and instance transformations to handle constraints more efficiently.

Findings

Achieved an EPTAS for scheduling with bag-constraints.

Developed a schedule repair method through polynomial-time job swapping.

Showed that respecting bag-constraints only among a constant number of bags suffices.

Abstract

Machine scheduling is a fundamental optimization problem in computer science. The task of scheduling a set of jobs on a given number of machines and minimizing the makespan is well studied and among other results, we know that EPTAS's for machine scheduling on identical machines exist. Das and Wiese initiated the research on a generalization of makespan minimization, that includes so called bag-constraints. In this variation of machine scheduling the given set of jobs is partitioned into subsets, so called bags. Given this partition a schedule is only considered feasible when on any machine there is at most one job from each bag. Das and Wiese showed that this variant of machine scheduling admits a PTAS. We will improve on this result by giving the first EPTAS for the machine scheduling problem with bag-constraints. We achieve this result by using new insights on this problem and restrictions given by the bag-constraints. We show that, to gain an approximate solution, we can relax the bag-constraints and ignore some of the restrictions. Our EPTAS uses a new instance transformation that will allow us to schedule large and small jobs independently of each other for a majority of bags. We also show that it is sufficient to respect the bag-constraint only among a constant number of bags, when scheduling large jobs. With these observations our algorithm will allow for some conflicts when computing a schedule and we show how to repair the schedule in polynomial-time by swapping certain jobs around.