Resource Management Through Workload Prediction Using Deep Learning in Fog-Cloud Architecture

Workload prediction involves forecasting the future resource demands and patterns for a computing system, such as cloud or fog infrastructure. It applies historical information and analytical methods to forecast future workload patterns. By anticipating workload patterns, resource managers can allocate resources proactively, optimize system performance, and ensure efficient resource utilization. This research introduces a Fog-Cloud specific workload prediction model based on time series analysis, utilizing Intuitionistic Fuzzified C-mean clustering and Long Short-Term Memory (IFCM-LSTM), with emphasis on CPU and memory prediction. First, IFCM clustering incorporates uncertainty, allowing for a more flexible representation of resource patterns. This is particularly useful in scenarios where resource utilization behavior is uncertain or hesitant. Next, the LSTM model effectively captures the intricate temporal relationships in the historical workload data collected from fog-cloud nodes. The model’s performance has been assessed on a real workload dataset using three evaluation metrics: Root Mean Square Error (RMSE), Mean Squared Error (MSE), and Mean Absolute Error (MAE). These metrics offer valuable insights into the accuracy of the model’s predictions.