Heterogeneous Computational Scheduling Using Adaptive Neural Hyper-Heuristic

IF 0.5 4区 数学 Q3 MATHEMATICS
A. Allahverdyan, A. Zhadan, I. Kondratov, O. Petrosian, A. Romanovskii, V. Kharin, Yin Li
{"title":"Heterogeneous Computational Scheduling Using Adaptive Neural Hyper-Heuristic","authors":"A. Allahverdyan,&nbsp;A. Zhadan,&nbsp;I. Kondratov,&nbsp;O. Petrosian,&nbsp;A. Romanovskii,&nbsp;V. Kharin,&nbsp;Yin Li","doi":"10.1134/S106456242460221X","DOIUrl":null,"url":null,"abstract":"<p>In heterogeneous computing environments, efficiently scheduling tasks, especially those forming Directed Acyclic Graphs (DAGs), is critical. This is particularly true for various Cloud and Edge computing tasks, as well as training Large Language Models (LLMs). This paper introduces a new scheduling approach using an Adaptive Neural Hyper-heuristic. By integrating a neural network trained with genetic algorithms, our method aims to minimize makespan. The approach uses a two-level algorithm: the first level prioritizes tasks using adaptive heuristic and the second level assigns resources based on the Earliest Finish Time (EFT) algorithm. Our tests show that this method significantly improves over traditional scheduling heuristics and other machine learning-based approaches. It reduces the makespan by 6.7% for small-scale DAGs and 28.49% for large-scale DAGs compared to the leading DONF algorithm. Additionally, it achieves a proximity of 84.08% to 96.43% to the optimal solutions found using Mixed-Integer Linear Programming (MILP), demonstrating its effectiveness in diverse computational settings.</p>","PeriodicalId":531,"journal":{"name":"Doklady Mathematics","volume":"110 1 supplement","pages":"S151 - S161"},"PeriodicalIF":0.5000,"publicationDate":"2025-03-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1134/S106456242460221X.pdf","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Doklady Mathematics","FirstCategoryId":"100","ListUrlMain":"https://link.springer.com/article/10.1134/S106456242460221X","RegionNum":4,"RegionCategory":"数学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q3","JCRName":"MATHEMATICS","Score":null,"Total":0}
引用次数: 0

Abstract

In heterogeneous computing environments, efficiently scheduling tasks, especially those forming Directed Acyclic Graphs (DAGs), is critical. This is particularly true for various Cloud and Edge computing tasks, as well as training Large Language Models (LLMs). This paper introduces a new scheduling approach using an Adaptive Neural Hyper-heuristic. By integrating a neural network trained with genetic algorithms, our method aims to minimize makespan. The approach uses a two-level algorithm: the first level prioritizes tasks using adaptive heuristic and the second level assigns resources based on the Earliest Finish Time (EFT) algorithm. Our tests show that this method significantly improves over traditional scheduling heuristics and other machine learning-based approaches. It reduces the makespan by 6.7% for small-scale DAGs and 28.49% for large-scale DAGs compared to the leading DONF algorithm. Additionally, it achieves a proximity of 84.08% to 96.43% to the optimal solutions found using Mixed-Integer Linear Programming (MILP), demonstrating its effectiveness in diverse computational settings.

在异构计算环境中,高效调度任务,尤其是那些形成有向无环图(DAG)的任务,至关重要。对于各种云计算和边缘计算任务以及大型语言模型(LLM)的训练而言,尤其如此。本文介绍了一种使用自适应神经超启发式的新调度方法。通过将经过遗传算法训练的神经网络整合在一起,我们的方法旨在最大限度地减少时间跨度。该方法使用两级算法:第一级使用自适应启发式确定任务的优先级,第二级根据最早完成时间(EFT)算法分配资源。我们的测试表明,与传统的调度启发式方法和其他基于机器学习的方法相比,这种方法有明显改善。与领先的 DONF 算法相比,小规模 DAG 的 makepan 降低了 6.7%,大规模 DAG 的 makespan 降低了 28.49%。此外,它还实现了 84.08% 至 96.43% 的接近度,接近于使用混合整数线性规划(MILP)找到的最优解,证明了它在各种计算环境中的有效性。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
求助全文
约1分钟内获得全文 求助全文
来源期刊
Doklady Mathematics
Doklady Mathematics 数学-数学
CiteScore
1.00
自引率
16.70%
发文量
39
审稿时长
3-6 weeks
期刊介绍: Doklady Mathematics is a journal of the Presidium of the Russian Academy of Sciences. It contains English translations of papers published in Doklady Akademii Nauk (Proceedings of the Russian Academy of Sciences), which was founded in 1933 and is published 36 times a year. Doklady Mathematics includes the materials from the following areas: mathematics, mathematical physics, computer science, control theory, and computers. It publishes brief scientific reports on previously unpublished significant new research in mathematics and its applications. The main contributors to the journal are Members of the RAS, Corresponding Members of the RAS, and scientists from the former Soviet Union and other foreign countries. Among the contributors are the outstanding Russian mathematicians.
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术官方微信