English

Efficient Algorithms for Scheduling Moldable Tasks

Data Structures and Algorithms 2023-03-30 v9

Abstract

We study the problem of scheduling nn independent moldable tasks on mm processors that arises in large-scale parallel computations. When tasks are monotonic, the best known result is a (32+ϵ)(\frac{3}{2}+\epsilon)-approximation algorithm for makespan minimization with a complexity linear in nn and polynomial in logm\log{m} and 1ϵ\frac{1}{\epsilon} where ϵ\epsilon is arbitrarily small. We propose a new perspective of the existing speedup models: the speedup of a task TjT_{j} is linear when the number pp of assigned processors is small (up to a threshold δj\delta_{j}) while it presents monotonicity when pp ranges in [δj,kj][\delta_{j}, k_{j}]; the bound kjk_{j} indicates an unacceptable overhead when parallelizing on too many processors. The generality of this model is proved to be between the classic monotonic and linear-speedup models. For any given integer δ5\delta\geq 5, let u=δ212u=\left\lceil \sqrt[2]{\delta} \right\rceil-1\geq 2. In this paper, we propose a 1θ(δ)(1+ϵ)\frac{1}{\theta(\delta)} (1+\epsilon)-approximation algorithm for makespan minimization where θ(δ)=u+1u+2(1km)\theta(\delta) = \frac{u+1}{u+2}\left( 1- \frac{k}{m} \right) (mkm\gg k). As a by-product, we also propose a θ(δ)\theta(\delta)-approximation algorithm for throughput maximization with a common deadline.

Keywords

Cite

@article{arxiv.1609.08588,
  title  = {Efficient Algorithms for Scheduling Moldable Tasks},
  author = {Xiaohu Wu and Patrick Loiseau},
  journal= {arXiv preprint arXiv:1609.08588},
  year   = {2023}
}

Comments

Accepted by European Journal of Operational Research