A multi-objective optimization technique for scheduling real-time IoT Applications in Fog Computing Using Approximate Computations and TOPSIS

Abstract

In the last decade, fog computing arose as a distributed computing paradigm to handle latency-sensitive real-time IoT applications in an effective way. By utilizing fog resources, improved performance such as timely service provisioning, optimal energy usage, decreased network load, etc. can be achieved. Fog resources usually have finite computational capacities. Conversely, Internet of Things (IoT) applications are getting complicated in addition to being computationally intensive, necessitating a specific degree of QoS in stringent time restrictions. In real-time, many times it is preferable for an IoT application to complete its execution by its deadline by generating an imprecise outcome instead of yielding a delayed accurate output. We study the placement of real-time IoT applications in a heterogeneous fog infrastructure by applying approximate computations. In this technique, we considered that if a constituent task yields an inaccurate outcome, the error may not only be limited to its closest predecessor tasks but may also proliferate to the succeeding workflow tasks, thereby, affecting the overall result of the workflow. We simultaneously study the impact of error proliferation on energy consumption of fog resources. The proposed workflow orchestrating model is compared to a baseline technique and a state-of-the-art policy, where the effects of partial computations are studied under varying values of proliferation probability of input error and result precision threshold. The simulation findings reveal that the proposed technique outperformed both policies in terms of the number of deadline misses, energy savings, schedule hole utilization, and overall result precision.