Treffer: Cerebrum: A Proactive Federated Learning Framework for Multi-Objective Edge Workflow Scheduling.

The rise of 5G networks and the Internet of Things (IoT) is driving a shift towards edge computing, where applications increasingly consist of complex, containerized workflows. Existing orchestration platforms, designed for centralized data centers, are reactive, workflow-agnostic, and poorly scalable, leading to high makespan, energy consumption, and cost. This paper presents Cerebrum, a proactive, learning-based framework for multi-objective scheduling of containerized workflows at the edge. Cerebrum combines several key innovations: (1) hierarchical workflow decomposition for scalable DAG partitioning, (2) a proactive multi-objective reinforcement learning agent with predictive resource forecasting, (3) a two-tiered architecture separating global inter-cluster orchestration from local intra-cluster scheduling, and (4) a unified policy framework balancing performance, energy, and cost. Extensive simulations show that Cerebrum achieves the lowest average makespan (90 s, 5.3% improvement over FL-Proactive), reduces energy consumption to 1,950 kJ (6.0% lower), and cuts monetary cost to $2.4 (36.8% lower), while maintaining scheduling decision latency below 15 ms across 20 clusters. These results demonstrate that proactive, adaptive scheduling intelligence can simultaneously optimize multiple objectives, providing a robust and scalable solution for next-generation edge-cloud applications. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Grid Computing is the property of Springer Nature and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)