An Effective Workload Prediction with Rnn-Lstm For Efficient Resource Autoscaling In Private Cloud Environments

Narek Badjajian^*1, Sandy Montajab Hazzouri²

¹ University of Debrecen, Department of Mathematical and Computational Science, Debrecen, Hungary

² Faculty of Informatics Engineering , Albaath University, Syria

Emails: badjajiann6math@gmail.com; Samonhaco1994@gmail.com

Abstract

The research focuses on an accurate workload prediction approach for auto-scaling resources in the Private Cloud using improved Time-Series models. Although many factors still result in dynamic workloads of cloud systems, an accurate forecast becomes vital for service quality and cost. The chapter discusses a Proactive Prediction Engine (PPE) framework using Auto Regressive Integrated Moving Average (ARIMA) and Recurrent Neural Network Long Short-Term, to forecast CPU utilization. Real-time datasets of OpenStack private cloud and Amazon AWS were used for experimental evaluation. The analyses show that the RNN_LSTM model performs far better than ARIMA by reducing the MAE and RMSE values by roughly 40 percent in each set. This has further reinforced that RNN_LSTM can model non-linearity and handle correlation issues in the workload data. Automated scaling of the instances with the Open Stack based on the predicted CPU load is made possible by the integration of RNN_LSTM prediction with OpenStack, supported by Terraform. This strategy reduces times of service outages and enables the efficient use of resources in the network. Regarding accuracy and automation, the proposed method can be a relevant solution for workload management for private cloud infrastructure. In this respect, the results support the implementation of deep learning-based predictive models to optimize the performance of autoscaling.

Keywords: Proactive Prediction Engine (PPE); Workload Prediction; ARIMA; RNN_LSTM; Hybrid Cloud; Deep Learning and Dynamic Autoscaling.