부산대학교 도서관

Nowadays, Cloud computing environment attracts many application developers to deploy their Web applications on cloud data centers. However, the Web applications workload continuously fluctuates depending on the user behavior, so it very hard to predict. Predicting future request workload is very important because it can help Cloud service providers automatically adjust resources online to maintain the Service Level Agreement (SLA). This paper proposes a Web application load prediction model by using Recurrent Neural Network (RNN) to forecast request workload in the future. The model uses a special type of RNN called Bi-directional Long Short-term Memory Network (Bi-LSTM). The experiment result is evaluated with real workload history from World Cup 1998 Website. Results show that our proposed model can improve around 50% prediction accuracy compared to regular LSTM and have an acceptable prediction speed for applying to real-time circumstances like the auto-scaling system.