Prediction of Amazon spot price based on chaos theory using ANFIS model

Abstract

Spot Instance Market is the most recent advancement in cloud computing business models. It is introduced by Amazon's Elastic Compute Cloud (Amazon EC2) in order to utilize its idle resources more efficiently. The main characteristic of spot instance is its dynamic pricing. The hourly price for a spot instance fluctuates depending on the supply and demand for cloud resources. Users across the globe can bid for a spot instance using an online auction platform. The auction platform determines the current market price, a.k.a. “Spot price” and the users whose bids are above the spot price obtain the instance. Amazon publicizes current spot price but does not disclose how it is determined. The major challenge for the users in this new business model is to predict the spot price before bidding. In this paper, we propose a new spot price forecasting model based on chaos theory. The proposed method makes use of chaos time series analysis to verify the chaotic feature of Amazon spot price and to perform a prediction using Adaptive Neural Fuzzy Inference System (ANFIS). We perform extensive simulation experiments using real spot price traces and show that the proposed method can be a bright merit to predict Amazon spot price.