Title: Amazon EC2 spot price prediction using LSTM time series prediction model

Authors: Khandelwal Veena; Khandelwal Shantanu

Addresses: Manipal University, Jaipur, India ' KPMG Services Pte. Ltd, Singapore

Abstract: Amazon EC2 spot instances provide access to unused Amazon EC2 capacity at high discounts relative to on-demand and reserved prices. Spot prices fluctuate based on the demand and supply of available unused capacity of EC2. When users request spot instances, they specify the maximum spot price they are willing to pay. Optimum maximum spot price estimation is crucial to control costs and have uninterrupted access to spot instances. We analyse spot price fluctuations for any seasonal or residual component and present a stacked LSTM-based prediction model based on the deep learning RNN model. In order to analyse Amazon spot pricing, we use time-smoothed spot prices at frequency of one hour. Our experiments with the new Amazon EC2 spot pricing model show that the LSTM model predicts future spot prices with different lead times with very low RMSE values.

Keywords: Amazon EC2; compute instances; new spot pricing model; spot price prediction; long short-term memory; LSTM.

DOI: 10.1504/IJCC.2024.136289

International Journal of Cloud Computing, 2024 Vol.13 No.1, pp.62 - 79

Received: 18 Jan 2022
Accepted: 04 Apr 2022
Published online: 26 Jan 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article