Particle Swarm Optimization (PSO) Long Short-Term Memory Network Based Cloud Resource Usage Prediction Model for Cloud Data center

Abstract

The deployments of cloud data centers are growing at an exponential rate. As the excessive use of computer resources and the accompanying energy and environmental impacts have a close correlation and therefore it is vital to be more concerned and careful about the emission of carbon footprints while improper usage of data center resources takes place. Virtual machine consolidation techniques are used by cloud service providers to efficiently manage computer resources and reduce energy consumption. Aggressive virtual machines (VM) consolidation, on the other hand, may result in service performance reduction and, as a result, serious violations of SLA. The most important challenge in appropriately managing the cloud resources such as memory and central processing units while limiting energy consumption is the accurate forecasting of Virtual Machine (VM) workloads. Various researches have been done to manage cloud resources and lowering resource costs through VM workload prediction using deep learning techniques. However, workload prediction accuracy still needs improvement. This research paper proposes a hybrid particle swarm optimization (PSO) and state-of-the-art long short-term memory (LSTM) deep learning model to solve this challenge. The proposed prediction model has been efficiently optimized using the PSO, allowing it to better forecast incoming workloads. To compare various metrics and extend the applicability of the model, the method first employs multivariate analysis to derive the sequential and contextual properties of historical workload data using the PSO based LSTM network from the original high-dimensional workload data. The model then combines the various hyperparameters (optimizers, metrics, and losses), and layers (Normalization, Recurrent, and Activation layers) to achieve accurate workload prediction. The experiment was done on a Bitbrain workload trace to show that the proposed method had a higher prediction accuracy. The Root Mean Squared Error (RMSE) findings show that the PSO-LSTM model beats state-of-the-art workload prediction approaches such as CNN, CNN-LSTM and LSTM in terms of VM workload forecasting accuracy in cloud computing environments up.