Dynamic deadline constrained multi-objective workflow scheduling in multi-cloud environments

作     者：Cai, Xingjuan Zhang, Yan Li, Mengxia Wu, Linjie Zhang, Wensheng Chen, Jinjun 

作者机构：Taiyuan Univ Sci & Technol Shanxi Key Lab Big Data Anal & Parallel Comp Taiyuan 030024 Shanxi Peoples R China Nanjing Univ State Key Lab Novel Software Technol Nanjing Peoples R China Chinese Acad Sci Inst Automat State Key Lab Management & Control Complex Syst Beijing Peoples R China Swinburne Univ Technol Dept Comp Sci & Software Engn Melbourne Australia 

出 版 物：《EXPERT SYSTEMS WITH APPLICATIONS》 (专家系统及其应用)

年 卷 期：2024年第258卷

核心收录：

学科分类：1201[管理学-管理科学与工程(可授管理学、工学学位)] 0808[工学-电气工程] 08[工学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：China University Industry-University-Research Collaborative Innovation Fund (Future Network Innovation Research and Application Project) [2021FNA04014] Key R&D program of Shanxi Province, China National Natural Science Foundation of China Central Government Guides Local Science and Technology Development Funds, China [YDZJSX2021A038] Open Fund of State Key Laboratory for Novel Software Technology (Nanjing University) , China [KFKT2022B18] 

主　　题：Multi-cloud Dynamically constrained multi-objective optimization Workflow scheduling Outage rates Deadline constraints 

摘      要：Workflow scheduling becomes difficult and demanding in multi-cloud systems because of the variety of billing models and resource kinds, as well as the susceptibility of processes to time limitations. In which the execution duration of the workflow needs to be flexibly adjusted according to the urgency or otherwise of the real situation, in this work, we model the workflow scheduling problem with deadline constraints as a dynamically constrained multi-objective optimization problem (DCMOP), where the disruption rate of the special cloud resources and the failure probability in different cloud environments are expressed as the disruption probability of the workflow in a comprehensive manner while considering the execution time and cost. Dynamics arise from price changes for specialized cloud resources and changes in the execution duration of workflows. In addition, an algorithm for dynamically constrained multi-objective optimization (DC-MOEADPDS) is proposed in this paper. The algorithm based on two-population synergy as well as diversity selection. The auxiliary population ignores constraint limitations to help the main population speed up convergence, and diversity selection enables the auxiliary population to have better diversity to assist the general public in looking for more viable areas. Through dynamic constrained workflow simulation experiments in a multi-cloud environment, our algorithm reduces the execution time by an average of 13.29%, the cost by an average of 32.26%, and the interruption probability by 56.75%. In addition, our algorithm outperforms other algorithms in experiments on a dynamically constrained benchmark set.