A decomposition-based multi-objective evolutionary algorithm with reinforcement learning for workflow scheduling in cloud computing environment

Cloud computing has become an integral part of modern computer science. Cloud service providers (CSPs) often have multiple conflicting objectives for different user requirements. Thus, workflow scheduling in cloud computing environment is a challenge multi-objective optimization problem (MOP). The m...

Full description

Saved in:
Bibliographic Details
Published in:Cluster computing Vol. 28; no. 10; p. 678
Main Authors: Xue, Fei, Wen, Jinbu, Wang, Peiwen, Fan, Wenyu, Geng, Yuge, Dong, Tingting
Format: Journal Article
Language:English
Published: New York Springer US 01.10.2025
Springer Nature B.V
Subjects:
Cloud computing
Completion time
Computer Communication Networks
Computer Science
Convergence
Decomposition
Efficiency
Evolutionary algorithms
Genetic algorithms
Heuristic
Machine learning
Multiple objective analysis
Operating Systems
Optimization
Optimization algorithms
Parameter sensitivity
Processor Architectures
Scheduling
User requirements
Workflow
Cloud computing
MOEA
Workflow scheduling
Reinforcement learning
ISSN:1386-7857, 1573-7543
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Cloud computing has become an integral part of modern computer science. Cloud service providers (CSPs) often have multiple conflicting objectives for different user requirements. Thus, workflow scheduling in cloud computing environment is a challenge multi-objective optimization problem (MOP). The multi-objective evolutionary algorithm (MOEA) is one of the most commonly used approachs, but it is sensitive to parameter settings and exsists the problems of early convergence and local optimum. To improve the convergence speed and optimality of the solution, the augmented Tchebychef (ATCH) as the objective decomposition method is adopted in the MOEA based on decomposition (MOEA/D). For the problem with sensitive parameter settings, Q-learning in reinforcement learning (RL) is designed to select the optional parameter adaptively in the ATCH method. This paper proposed a decomposition-based MOEA with Q-learning (QLMOEA/D) to solve the multi-objective workflow scheduling problem while taking into account the task completion time (makespan), cost and load. Experimental results demonstrate that the proposed QLMOEA/D achieves better convergence and diversity on both benchmark functions (ZDT and DTLZ) and real-world scientific workflows (SWFs). It obtains the best performance in 55.56% of all test cases and outperforms baseline algorithms in 94.44% of scenarios across makespan, cost, and load objectives.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1386-7857
1573-7543
DOI:10.1007/s10586-025-05369-y