Optimal prediction of cloud spot instance price utilizing deep learning

Abstract

Cloud platforms frequently provide a variety of virtual machine models (VMs) with varying types and capacities, allowing users to select instances that best suit their needs. Cloud providers have devised a system for maximizing the utilization of redundant computing resources, whose costs fluctuate dynamically depending on supply and demand. "Spot pricing" is a common term for this, the user must make a suitable offer that is higher than the spot price to use this instance. Accurate spot price prediction enables users to prepare ahead of time for the bid price and run time to increase the method's reliability. To solve this critical challenge of predicting future pricing, we utilize a consistent and robust deep learning model. When compared to other sophisticated methods, the test results reveal that our proposed method performs better and more accurately.

Applying Machine Learning in Cloud Service Price Prediction: The Case of Amazon IaaS

“…Spot instance price prediction was carried out using k-nearest neighbors (KNN) in [11]. In [12,13], the authors adopted several machine-learning approaches to forecast the future pricing of EC2 spot instances, such as linear, ridge, and lasso regressions, multilayer perceptrons, k-nearest neighbors, extra trees, and random forests. Based on the results, machine learning is an encouraging approach for predicting spot prices.…”

“…In order to provide a comprehensive comparison, several studies have examined spot instance pricing and reported MAPE values. These studies [11][12][13] have found that shortterm forecasts, typically ranging from 1 to 3 hours ahead, exhibit MAPE values between 10% and 20%. Conversely, longer-term forecasts, specifically those made 24 h in advance, demonstrate MAPE values ranging from 20% to 40%.…”

“…Second, the data can be fed to machine-learning models capable of encompassing complicated patterns and making accurate predictions [9]. Current work focuses on how to use machine-learning techniques to predict spot instances prices to enhance the biting process [10][11][12][13]. In this case, prediction is based on data collected a few hours earlier.…”

Evaluating Amazon EC2 Spot Price Prediction Models Using Regression Error Characteristic Curve

“…In [40], the authors proposed a modified gated recurrent unit (MGRU) model and compared its performance to five other statistical and deep learning methods:…”