Multivariate Deep Learning Model For Workload Prediction In Cloud Computing

Many application developers are now choosing to install their Web applications on cloud data centers because of the attractiveness of cloud computing environment. Predicting future resource workload is critical since it allows cloud service providers to automatically modify resources online in order to meet service level agreements (SLA). This paper proposes a multivariate deep learning prediction model to predict future resource workload for cloud computing environment. The prediction model uses a special type of recurrent neural network (RNN) called Bidirectional long short-term memory (Bi-LSTM). This work also explains and shows the advantage by using multivariate data compared to univariate data in time series forecasting. The experiments, using real world workload dataset, show that the proposed multivariate Bi-LSTM model outperforms the univariate Bi-LSTM model in prediction accuracy.