Adaptive Virtual Clustering Methods for Dynamic IoT Edge Systems

Abstract

The majority of traditional clustering mechanisms are not competent to resource elasticity over multiple IoT systems and hence, handling load distribution effectively underutilizes the resources. Adapting to the complexities and constraints imposed by these system requires adaptive mechanisms that utilize virtual clustering to improve performance and resource utilization. The article proposes the design and performance evaluation of adaptive virtual clustering methods tailored to the requirements of emerging IoT edge systems with dynamic environments. The goal is to optimize resource allocation, balance load, and enhance the overall system performance by deploying virtual clusters with the potential of adapting the best to changing network loads as well as device heterogeneity. Was developed a multistep approach that combined state-of-the-art clustering algorithms, the K-means and agglomerative clustering, to our insight of Amdahl's law using metalearning strategies. This study revisits KPIs such as Silhouette Coefficient, Davies-Bouldin Index and Calinski-Harabasz Index under different network loads and performance of the device being considered. The proposed methods achieved better performance than existing clustering algorithms, particularly in high network traffic scenarios. In competitive testing, adaptive virtual clusters with up to 10% better in total performance over traditional clusters at full network load through smart node allocation and leveraging of the virtual memory for better load balancing. Adaptive virtual clustering appears to be a promising fit for the challenges posed by dynamic IoT edge environments. This limber approach yields improved in network efficiency, load distribution and overall network performance. Future works can focus on the optimization of these approaches in different use cases and study how real-time adaptive mechanisms can be combined to enhance system responsiveness and efficiency.