A Deep Learning Approach for Amazon EC2 Spot Price Prediction

Abstract

Spot Instances (SI) represent one of the ways cloud service providers use to deal with idle resources in off-peak periods, where these resources are being auctioned at low prices to customers with limited budgets in a dynamic manner. However, SI are poorly utilized due to issues like out-of-bid failures and bidding complexity. Thus, effective SI price models are of great importance to customers in order to plan their bidding strategies. This paper proposes a deep learning approach for Amazon EC2 SI price prediction, which is a time-series analysis (TSA) problem. The proposed Long Short-Term Memory (LSTM) approach is compared with a well-known classical (i.e., non deep learning) approach for TSA, which is AutoRegressive Integrated Moving Average (ARIMA), using different accuracy measures commonly used in TSA. The results show the superiority of the LSTM approach compared with the ARIMA approach in many aspects.