.png){kind=icon}
.png){kind=icon}
Return
Reinforcement learning-assisted evolutionary auxiliary optimization for multi-workflow scheduling in clouds
.png){kind=icon}
PRE
AI
DOI：10.1016/j.swevo.2026.102422.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Cloud computing has become an important platform for the dramatic migration of workflow applications and the diversification of application requirements. However, multi-workflow scheduling is extremely challenging in the cloud as users may request various applications with different quality of service (QoS) demands. To address this challenge, this paper formulates the multi-workflow scheduling as an optimization problem with an auxiliary task (OPAT) by exploiting the relationship between workflows, and proposes a Reinforcement Learning-assisted Evolutionary Auxiliary Optimization (RLEAO) approach to minimize the makespan, cost, and energy consumption of workflow execution while meeting QoS constraints. Specifically, a correlation mining strategy is designed to explore the potential synergistic benefits between workflows, so as to identify an auxiliary optimization problem (task) and construct the OPAT. Further, a reinforcement learning-guided knowledge transfer strategy is developed to automatically transfer diversified knowledge between the two optimization tasks while avoiding negative transfer. Lastly, an adaptive population regulation strategy is introduced, which dynamically allocates solutions based on feedback during population evolution, thereby improving the search performance of the algorithm in the main optimization task. Extensive experiments with various real-world workflow applications demonstrate that RLEAO outperforms some state-of-the-art methods by providing a better set of trade-off solutions for multi-workflow scheduling in a multi-cloud system.
Keywords:
multi-workflow scheduling
cloud computing
reinforcement learning
evolutionary optimization
quality of service
Journal
IF:
8.5
Papers:
2.1K
Citations:
1.0W