Intelligent Task Offloading for Latency-Sensitive Fog-IoT Architectures

Abstract

Latency-sensitive IoT applications need task execution in the proximity of data sources under resource constraints at devices, varying bandwidth in communication, and dynamic congestion at fog nodes. Typical offloading techniques usually analyze either the transmission, computation, queuing, or deadline individually, leading to the choice of a fog node that may look attractive in terms of computing power but turns out to be inappropriate as the queue at the fog node approaches saturation. In this paper, we propose an Intelligent Queue-aware Task Offloading (LRI-IQATO) approach to Latency-Sensitive Fog-IoT systems. The proposed approach includes the construction of a nonlinear function called Queue-Aware Latency Risk Index (LRI), where the risk of transmission, propagation, processing, and congestion is incorporated into the latency, which is then normalized by the deadline of the task. The degree of the risk penalty as a function of congestion sensitivity is included in the LRI formulation. From the LRI constraint, a closed-form task-specific maximum fog-utilization threshold is derived to rule out infeasible fog nodes before offloading the task.Three operating regime of three regimes are used to classify execution states as being low risk, transition risk, and high risk. Numerical analysis of the approach has been done to showcase the behaviour of the proposed model. The approach is considered to be a framework for decision making that will support the intelligent offloading policy and not a new queuing model or machine learning technique.