LSRU: A Novel Deep Learning based Hybrid Method to Predict the Workload of Virtual Machines in Cloud Data Center
Md. Nahid Hasan Shuvo, Md. Nahid Hasan Shuvo, Mirza Mohd Shahriar Maswood, Mirza Mohd Shahriar Maswood, Abdullah G. Alharbi, Abdullah G. Alharbi
Abstract
Task scheduling is a key innovation of cloud computing. However, forecasting resource usage depends on the previous usage of resource. Thus, this type of problem is modeled as a time series prediction problem. Cloud computing platform gives the ability of sharing both hardware and software resources on demand. Because of dynamic change of resource request in data center, it is difficult to predict the future workload precisely. Thus, we propose a novel hybrid-method for improving the accuracy, named LSRU, which is mainly the combination of two models i.e. Gated Recurrent Unit (GRU) and Long Short Term Memory (LSTM). Our method LSRU achieves better accuracy in comparison with different statistical models, and LSTM and GRU model separately. LSRU can predict short-time ahead prediction along with long-time ahead prediction with sudden burst of workload. Experimental results show the performance of LSRU compared to other existing models and the visual representation of forecasting workload traces of different types and duration.