NVM-Enhanced MLI Placement for Revenue Maximization in UAV-Fog Assisted MEC with Stable Matching

Abstract

The rise of smart edge devices and the growing demand for advanced technologies such as Machine Learning (ML) necessitate an evolution beyond 5G networks. Placing Machine Learning Inference Instances (MLIs) at the edge server can reduce latency but faces memory and processing constraints. By combining Mobile Edge Computing (MEC) and Non-Volatile Memory (NVM), Service Providers (SP) can efficiently deploy MLIs on edge, fog, and cloud servers to reduce latency and improve quality by offloading compute-intensive tasks. To serve large geographies, Unmanned Aerial Vehicles (UAVs) can be leveraged in large-scale sparsely distributed edge, fog, and cloud systems. This paper explores a 3-tier edge-fog architecture with NVM memory and introduces a placement scheme to maximize SP revenue by strategically deploying MLIs on UAV and fog with NVM technology using a stable matching-based method. © 2024 ACM.