Content-Specific and Buffer-Based Migration Schemes for Fog Computing

Abstract

Fog computing and network function virtualization (NFV) technologies reduce latency and provide scalable services at the network edge. However, the limited resources of edge nodes pose a challenge in handling high traffic volumes from requests that demand high computation, extended lifetime, and low latency. To address this challenge, dynamic load migration schemes for NFV-based fog paradigms are proposed here. First, a content-specific scheme that diffuses excess loads to nearby locations hosting relevant virtual network functions (VNFs) of the requests. Second, a buffer scheme reserves a dedicated node to absorb loads from saturated nodes in the proximity of terminals. Third, a hybrid scheme integrates both strategies by initially migrating loads to the pre-allocated buffer to provide immediate relief without searching for candidate nodes. Upon the buffer saturation, it switches to the content-specific phase to distribute loads to the proximate nodes. The network introduces a novel request model comprised of dependent and independent VNFs of varying resource demands. Dependent VNFs are collectively mapped on a primary node while distributing independent VNFs across neighboring secondary nodes, forming a structured ring topology mapping method. These schemes enhance migration success rates, and reduce migration iterations, service downtime, and cost, as compared to prominent solutions.