.png){kind=icon}
.png){kind=icon}
Return
A three-tier AI-based approach for dynamic application placement in cloud-edge environments
.png){kind=icon}
PRE
AI
DOI：10.1016/j.future.2026.108511.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The transition to the 6G era and the rapid proliferation of Artificial Intelligence (AI)-driven functionalities introduce a key challenge in edge-cloud environments: enabling dynamic orchestration and adaptive allocation of compute and network resources while satisfying end-user latency requirements. Addressing this challenge requires the cooperative and dynamic resolution of multiple factors to support the deployment and management of complex, time-critical edge applications. This involves balancing compute, network, and storage resource allocation during both deployment and runtime, while ensuring placement decisions adapt to evolving user demands. To address the increasing data volume and AI-driven workload growth, we propose a threetier AI-based orchestrator. The system proactively characterizes system behavior using application-specific, infrastructure-level, and user-centric information to determine optimal edge-application placement. Placement decisions are further enhanced using real-time resource utilization metrics and dynamic user proximity to edge nodes. The architecture follows a distributed design. The first two tiers employ Incremental Learning and Federated Learning, respectively, while the final tier applies centralized Reinforcement Learning to determine final placement actions. This design enables adaptive and efficient workload distribution while balancing resource utilization and user demand. The proposed workload placement approach is validated across 100 simulated scenarios with dynamic user proximity and fluctuating edge resources, achieving 97.0% accuracy, defined as the proportion of placements that satisfy resource constraints while preserving user proximity.
Keywords:
Edge computing
Cloud computing
Orchestration
Incremental learning
Federated learning
Reinforcement learning
Journal
F
IF:
6.1
Papers:
6.8K
Citations:
2.3W