A Deep Learning Model for Accurate Prediction in Cloud Computing Workloads

Abstract

Efficient resource allocation in Cloud computing necessitates the capability to accurately predict and adapt to Cloud workloads with both precision and flexibility. Nevertheless, current approaches struggle to predict Cloud workloads with high dimensionality. This leads to inefficient resource utilization and the failure to meet service level agreements (SLAs). For these important challenges to be addressed, a dynamic recurrent neural network is proposed to accurately predict workloads. The proposed algorithm integrates an auto-encoder to effectively extract representations from the original workload data with high dimensionality to enable adaptive and accurate predictions for highly variable workloads. Experiments are conducted using Google Cloud data centers to validate the effectiveness and adaptability of the proposed scheme across various workload types and prediction lengths. The proposed scheme achieves 81.48% lower MSE and maintains an MSE of 0.005 to 0.015 compared to traditional Deep learning-based methods with up to 0.027 for Cloud workloads. This is achieved by successfully predicting continuous resource metrics (CPU/Memory/GPU utilization) rather than categorical labels, demonstrating strong adaptability across various prediction lengths.