EA-FedPPO: Energy-Adaptive Federated Policy Learning for QoS-Aware Task Offloading in Multi-Tier Edge–Cloud Systems

Multi-tier edge–regional–cloud infrastructures exhibit inherent asymmetries in computational resources, communication conditions, and workload characteristics, creating significant challenges for energy-efficient and quality-of-service (QoS)-aware task offloading. Addressing these challenges requires coordinating intelligent task placement with efficient federated-learning (FL) participation. This paper proposes EA-FedPPO, an energy-adaptive federated proximal policy optimization framework that integrates multi-criteria client selection with hierarchical, QoS-aware task scheduling. The proposed framework introduces a client-selection mechanism that combines historical PPO-update magnitude per unit of federated energy, client staleness, and participation diversity to guide training participation while controlling federated overhead. At the scheduling level, a global PPO policy selects execution tiers using task characteristics and dynamic system conditions, followed by queue-aware node selection based on projected completion delay. Through this coordinated architecture, EA-FedPPO accounts for resource, communication, and workload asymmetries while balancing energy consumption, scheduling delay, and service-level agreement (SLA) compliance. The framework is evaluated against four federated participation strategies under typical and elevated loads, with workloads ranging from 1000 to 10,000 tasks. Under typical load at 10,000 tasks, EA-FedPPO achieves descriptive mean-delay reductions of 8.7–17.4% relative to the federated baselines, alongside the lowest mean SLA violation rate among these methods. Compared with full participation, EA-FedPPO reduces modeled federated-learning energy consumption by 23.75% through selective client participation, while maintaining comparable overall energy consumption. Additional experiments investigate client-selection ablation, reward sensitivity, and federation scalability under varying participation ratios and non-IID workloads. The findings demonstrate the potential of coupling energy-aware federated participation with QoS-aware scheduling, while revealing performance dependencies on workload intensity and system heterogeneity. Overall, EA-FedPPO provides an integrated approach to managing federated-learning overhead and scheduling trade-offs in asymmetric multi-tier edge–cloud environments.

Authors

Institutions

Publication Details

Journal
Symmetry
Published
2026-11-16
DOI
https://doi.org/10.3390/sym18111918
Primary Topic
IoT and Edge/Fog Computing
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

EA-FedPPO: Energy-Adaptive Federated Policy Learning for QoS-Aware Task Offloading in Multi-Tier Edge–Cloud Systems

Amirah H. Alomari
5 links
Symmetry
IoT and Edge/Fog Computing
article

EA-FedPPO: Energy-Adaptive Federated Policy Learning for QoS-Aware Task Offloading in Multi-Tier Edge–Cloud Systems

Amirah H. Alomari
article en

Abstract

Multi-tier edge–regional–cloud infrastructures exhibit inherent asymmetries in computational resources, communication conditions, and workload characteristics, creating significant challenges for energy-efficient and quality-of-service (QoS)-aware task offloading. Addressing these challenges requires coordinating intelligent task placement with efficient federated-learning (FL) participation. This paper proposes EA-FedPPO, an energy-adaptive federated proximal policy optimization framework that integrates multi-criteria client selection with hierarchical, QoS-aware task scheduling. The proposed framework introduces a client-selection mechanism that combines historical PPO-update magnitude per unit of federated energy, client staleness, and participation diversity to guide training participation while controlling federated overhead. At the scheduling level, a global PPO policy selects execution tiers using task characteristics and dynamic system conditions, followed by queue-aware node selection based on projected completion delay. Through this coordinated architecture, EA-FedPPO accounts for resource, communication, and workload asymmetries while balancing energy consumption, scheduling delay, and service-level agreement (SLA) compliance. The framework is evaluated against four federated participation strategies under typical and elevated loads, with workloads ranging from 1000 to 10,000 tasks. Under typical load at 10,000 tasks, EA-FedPPO achieves descriptive mean-delay reductions of 8.7–17.4% relative to the federated baselines, alongside the lowest mean SLA violation rate among these methods. Compared with full participation, EA-FedPPO reduces modeled federated-learning energy consumption by 23.75% through selective client participation, while maintaining comparable overall energy consumption. Additional experiments investigate client-selection ablation, reward sensitivity, and federation scalability under varying participation ratios and non-IID workloads. The findings demonstrate the potential of coupling energy-aware federated participation with QoS-aware scheduling, while revealing performance dependencies on workload intensity and system heterogeneity. Overall, EA-FedPPO provides an integrated approach to managing federated-learning overhead and scheduling trade-offs in asymmetric multi-tier edge–cloud environments.

SymmetryVol. 18(11)
King Khalid University (SA)
Industry, innovation and infrastructure
IoT and Edge/Fog Computing
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon