An Optimized Approach for Energy Consumption of Smart Devices in Fog Computing using Computational Intelligence Techniques

Abstract

To address the frequent overloading of fog nodes due to the increasing demand for IoT applications, an ensemble approach was employed to classify host load status into underloaded, balanced and overloaded categories. This work introduces an innovative reliability framework that encompasses multiple implementation phases. The process begins with the generation of virtual machines via the command line using various random settings.

Various parameters such as CPU utilization, number of CPU cores, RAM, memory allocation, memory availability, disk I/O, and network I/O were analyzed to better understand host workload. Three case studies with varying numbers of virtual machines (VMs) were conducted on two different platforms for load prediction.

A total of ten machine learning models were employed to construct an ensemble model, which ultimately yielded optimal and accurate results for classifying host load. All models are evaluated for their precision, recall, and accuracy. Various pre-processing techniques such as normalization, transformation, principal component analysis (PCA), outlier removal are applied on the generated dataset and various models are compared.

It was observed that applying normalization to a dataset improved the performance of the models. Four models—Random Forest (RF), AdaBoost (AB), Gradient Boost (GB), and Decision Tree (DT)— performed equally well across all three case studies with normalized datasets. However, our proposed ensemble model performed marginally better than these individual models and it achieved nearly 82% accuracy in correctly classifying host load.

As the next major revolution in cloud and fog computing environment, container migration and containerization have emerged as key advancement. Fog computing and mobile edge cloud necessitate the transfer of containers from overloaded hosts to new hosts to ensure adequate resources for executing consumer applications at the network edges. Despite the growing popularity of containers, algorithms to manage the excessive energy consumption of hosts have not been thoroughly investigated. Moreover, optimizing the energy consumption efficiency of hosts remains a critical and challenging task.

In order to address the critical issue of reducing energy consumption in fog computing environment, the study moved beyond traditional virtualization techniques, which have a high computational overhead and are less suitable for fog devices. Containers, known for their efficiency in encapsulating fog services, were used instead. A container selection algorithm was introduced to identify containers for migration when a host becomes overloaded.

Further, an energy-efficient container migration strategy was implemented using a dynamic inertia weight-based particle swarm optimization (DIWPSO) algorithm. This strategy aimed to balance the load and reduce energy consumption by migrating containers from overloaded hosts. Experimental results demonstrated that the DIWPSO algorithm significantly reduced energy consumption by 10.89% and achieved load balancing at a lower migration cost compared to traditional meta-heuristic solutions such as PSO, ABC, and E-ABC.

Additionally, The study developed a multivariate time series ensemble model for load prediction on hosts, utilizing anomaly detection techniques to forecast CPU utilization in the near future. Based on these predictions, resource utilization for container management was forecasted, determining the number of hosts needed to support the load of running containers.

Anomaly detection techniques were employed to reduce redundancy in generated data and address inconsistencies in load prediction due to the large volume of data. A predictive model with variable load patterns can better estimate future resource needs, which is crucial for capacity planning, meeting service-level goals, and achieving energy efficiency.

Various time series-based models were used for workload prediction, and the top three models were selected based on their TOPSIS scores to develop the ensemble model. To ensure the efficiency of the proposed model, Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), and accuracy were evaluated and compared with other existing stateof-the-art models. The results demonstrated that the proposed ensemble model exhibited higher accuracy in workload prediction compared to current state-of-the-art models, achieving the lowest Mean Absolute Percentage Error (MAPE) and providing an accuracy of approximately 88%.

In conclusion, by integrating advanced machine-learning models for load prediction with an optimized container migration strategy, the study effectively enhanced resource utilization and energy efficiency in fog computing environment. This comprehensive approach successfully addressed the dual challenges of load balancing and energy consumption, providing a robust solution for managing the increasing demands of IoT applications.