Local Load Migration in High-Capacity Fog Computing

Abstract

Fog computing brings storage and computational capabilities closer to the data source, which reduces latency and enhances efficiency in processing data. However, these capabilities are resource-constrained at the fog nodes as compared to the cloud core. This limitation can result in high computational loads that yield in saturation and congestion at the fog nodes when processing large traffic volumes. Hence, this paper introduces a novel load migration scheme for delay-sensitive and computation-intensive requests in fog computing without relaying to the cloud core, thus avoiding prolonged link delays. First, a grouped service function chain (SFC) provisioning framework is embedded on a heterogeneous architecture composed of super fog (SF) and ordinary fog (OF) nodes, where all the virtual network functions (VNFs) in the SFC are mapped on a single SF node to reduce resource use. Here, the SF nodes are always in active mode to receive incoming traffic, whereas the OF nodes are maintained in idle mode to save power and computing resources. When the SF node approaches a saturation threshold alarm, it diffuses its load (hosted VNFs) gradually to neighboring OF nodes in the vicinity, thus avoiding saturation at the SF node and providing service continuity. The framework is implemented on a resource-constrained network to achieve realistic operating conditions. Overall, the proposed work achieves high admission and resource utilization rates, reduced delays, and power consumption, as compared to key network architectures and provisioning schemes that separately map delay-sensitive and computation-intensive on hierarchical fog layers, which incur increased delays, network traffic, and power usage.