Adaptive Multi-Objective Energy-Efficient Resource Allocation Strategy for Cloud-Fog Environment for Smart Waste Monitoring System

Abstract

Internet of Things (IoT) based smart waste bin systems in urban environments have revived the interest in processing the waste data with low latency and cost driven by the concept of Cloud-Fog computing architectures. Existing strategies on road fog resource allocation are mainly aimed at minimising response time (and cost), while not prioritising energy as a key concern, limiting the viability of the strategies at city level. This paper presents an Adaptive Multi-Objective Energy-Efficient Resource Allocation (AMOEERA) strategy for Cloud-Fog based smart waste monitoring, which optimizes response time (time delay), energy and cost together, based on an adaptive scheduling model of urgency-aware scheduling, under a five-layer IoT-Fog-Cloud architecture (IoT/Edge, Fog, Proxy/Gateway, Cloud, and Management/Application). The performance of the AMOEERA is compared with the First-Come-First-Served (FCFS), Particle Swarm Optimization (PSO) and Non-dominated Sorting Genetic Algorithm II (NSGA-II) on the following parameters: Response Time, Energy, Cost, Resource Utilization and Throughput. The strategies were tested in five device-scaling configurations (50-500 end devices) for 12 seeds of each strategy under a discrete-event simulation parameterized following an iFogSim-based fog/cloud model; the differences between the strategies were evaluated via the paired Wilcoxon signed-rank test. Results show that the AMOEERA algorithm can generate an average of 9.9% energy saving and 26.0% increment of resource utilization and 12.1% increment of throughput (which are statistically significant p < 0.01) with respect to NSGA-II algorithm in most of the configurations. The response time is comparable to NSGA-II with a marginal improvement in the former at smaller scales and the cost premium is small and largest at smaller scales but closing rapidly towards equality at 500 devices, the scheduling overhead is 241-370 times slower, decreasing at larger scales. The results also show the benefit of the lightweight, urgency-aware, greedy scheduler over a population-based optimizer with respect to response time and cost at scale, as well as energy efficiency/energy usage, utilization and throughput at a relatively low level of overhead; beneficial for, say, a city-wide deployment of smart waste monitoring, which would likely be large but not massive (as in 500 devices were tested here, and the observed trends suggest that those advantages would remain at larger scale/deployments).