Asynchronous Coarse-Grained Load Migration Scheme for IoT Applications in Fog Networks

Fog computing provides distributed processing and storage solutions for real-time applications that demand low latency and fast response times. This makes fog solutions suitable for internet-of-things (IoT) devices that request various network functions. To offer multiple functions for IoT applications, fog nodes can leverage network function virtualization (NFV) for a scalable and elastic function modification, i.e., without the need for dedicated hardware. Despite the saliencies achieved from the synergistic combination between fog and NFV technologies, a key challenge here is the limited resources at the fog nodes. This makes the latter susceptible to rapid node saturation and network congestion at high traffic volumes. Along this, efficient resource utilization and load distribution mechanisms are necessary to enhance admission rates and quality-of-service (QoS). Hence, this paper proposes novel load migration schemes for NFV-based fog networks that aim to reduce the overhead of the migration process. Namely, a coarse-grained load diffusion scheme is adopted to reduce migration frequencies, incurred delay, and cost. Further, destination nodes are selected based on least-load (LL) or least-delay (LD) mechanisms to accommodate delay-sensitive, delay-tolerant, and computation-intensive IoT applications.

I. Introduction

F og computing first proposed in [1] brings cloud services to the edge of the network at a closer proximity to the source of data (end-user terminals). This allows local data processing at reduced latency levels without sending them back to the cloud core. Along this, fog paradigms serve delay-sensitive IoT applications that demand rapid service instantiation at minimal control-plane delays. The motivation for fog computing is to reduce the aggregated delays in cloud domains due to the extended propagation distances over geographically dispersed datacenters. Fog computing provides reduced link delays, thus reducing latency and improving response times. It also enhances services for remote locations to enable data computation and storage, where access to the cloud core is limited. Further, the local processing of data reduces cost of bandwidth and alleviates traffic at the cloud domain. However, one limitation of fog computing is the limited resources. Consequently, fog nodes are vulnerable to network congestion and rapid saturation during high traffic volumes, particularly when terminals demand intensive data computation for prolonged durations. This exhausts network resources, degrades the QoS, and reduces capacity and revenue. Hence, efficient resource management is vital for fog computing for enhanced network utilization.