Amazon Spot Price Prediction using Bayesian Model

Amazon Spot Price Prediction using Bayesian Model

Cloud computing is emerging as a promising field for providing a variety of computing services. Each service is priced using different pricing schemes. Increased demand for public cloud resources has created significant interactions between price, efficiency, and reliability. Increased demand for public cloud resources has created significant interactions between price, efficiency and reliability. With spot instances, Amazon has offered cheap resources without reliability on resources. In general, these instances reduce the monetary costs for cloud users, but its pricing features have not been discovered and reported yet. In this research, first, we perform a comprehensive analysis on the spot instances based on a 10-month history of prices. Then, based on the Bayesian method, a model is suggested for predicting the next hour price. The results of the modeling show that the average of the differences between the prices obtained by the model and the actual prices is small and insignificant. The Bayesian method creates more ordered resulting prices, and in cases where the original prices have outlier instances (possibly due to noise or fluctuations in supply and demand), prevents a sudden increase or decrease in prices. That fact that the price resulting from the proposed model is close to the original prices shows that unlike previous research results, it is most likely that Amazon uses a reasonable pricing method and its pricing approach is not unfair.

Cloud computing is emerging as a promising field for providing a variety of computing services. Each service is priced using different pricing schemes. Increased demand for public cloud resources has created significant interactions between price, efficiency, and reliability. Increased demand for public cloud resources has created significant interactions between price, efficiency and reliability. With spot instances, Amazon has offered cheap resources without reliability on resources. In general, these instances reduce the monetary costs for cloud users, but its pricing features have not been discovered and reported yet. In this research, first, we perform a comprehensive analysis on the spot instances based on a 10-month history of prices. Then, based on the Bayesian method, a model is suggested for predicting the next hour price. The results of the modeling show that the average of the differences between the prices obtained by the model and the actual prices is small and insignificant. The Bayesian method creates more ordered resulting prices, and in cases where the original prices have outlier instances (possibly due to noise or fluctuations in supply and demand), prevents a sudden increase or decrease in prices. That fact that the price resulting from the proposed model is close to the original prices shows that unlike previous research results, it is most likely that Amazon uses a reasonable pricing method and its pricing approach is not unfair.