Workload Prediction using ARIMA Statistical Model and Long Short-Term Memory Recurrent Neural Networks

In recent times there is a growing demand for cloud-based applications (Software as a Service model) deployed on public or private clouds. Companies providing these services are competing and striving to provide the end-user a good quality of service (QoS), in order to attract the customers and increase the revenue over the time. But meeting the sufficient levels of QoS expected by the end-users, with cost-effective and manageable amount of resources, is often difficult, because the number of application requests on servers undergo variation in time. An efficient solution to this problem is proactive dynamic provisioning of resources, by estimating the future workload on the servers and allocate or release the resources as per requirement. In the present paper a workload prediction model using Long Short Term Memory (LSTM) Recurrent Neural Network (RNN), which is capable of remembering long-term dependencies, is proposed. The proposed LSTM-RNN model is compared with the Auto Regressive Integrated Moving Average (ARIMA) model and the accuracy of predictions obtained for LSTM-RNN model, is observed to be better than that of ARIMA model.