Cloud Workload Prediction Using ConvNet And Stacked LSTM

Today, with massive growth of cloud computing in recent years, service level agreement (SLA) and dynamic resource scaling for better services is of great importance. Investigating the cloud trace in order to have a prediction of traffic in future times for computational tasks is of great popularity in previous works. To solve this issue many efficient manners have been used. In this paper, we combined 1D ConvNets and stack of long-short term memory (LSTM) blocks to process long sequence of Google trace data in order to have a precise and light computing prediction of RAM and CPU requests in future timestamps. Experimental results confirms that our approach while having high accuracy, also not involved in heavy calculations and efficiently works with long sequences.