Forecasting the price of AWS On-spot instances using Deep Neural Network Architectures

No Thumbnail Available
Jaishankar, Ranjith Kumar
Issue Date
MSc in Data Analytics
Dublin Business School
Items in eSource are protected by copyright. Previously published items are made available in accordance with the copyright policy of the publisher/copyright holder.
Among the cloud computing services, the concept of On-spot instance is the most popular which has been introduced by Amazon AWS, in order to utilize their spare capacity. On-spot instance follows the auction-based cloud model, where the price on-spot changes with time. In general analysis, it has been found that AWS On-spot instances are 30-40% cheaper than regular instances. The concept of dynamic pricing for AWS on-spot instance makes it complicated for some users, to bid for an optimal price. In order to help the users for selecting the optimal price with AWS On-spot instances, this research is predicted using the 4 different deep learning architectures which includes CNN, RNN, LSTM and Bi-LSTM for price prediction. To select the best performing model MSE, RMSE and MAE score has been calculated for each model over the test data. The better outcomes is achieved using Bi-LSTM model in terms of performance. In order to implement the concept of ON-spot price prediction, a web-portal using python flask has been developed which provides the predicted price of On-spot instance based on user input such as Region, Operating system, instance type, Time stamp etc.