MACHINE LEARNING-BASED VM PERFORMANCE PREDICTION IN AWS CLOUD USING GRU AND DTW TECHNIQUES

Abstract

Cloud computing, such as AWS cloud, is vital for solving the growing needs of applications demanding to calculate economic computing and storage resources. Since the dependence on cloud services such as AWS is increasing, it is necessary to optimize the allocation of cloud resources. Cloudprophet introduces a new machine learning methodology to predict the performance of virtual machines (VMS) in cloud settings. This approach uses Dynamic Time Warping (DTW) to classify application types and uses Pearson's correlation to detect strongly connected Runtime metrics. Measurements are used in three variants of machine learning algorithms: LSTM without highly selected and DTW metrics, LSTM with highly selected and DTW metrics and GRU with DTW and highly correlated metrics. The GRU, including both DTW and highly correlated measures, overcomes others, with an accuracy of 99.3% when predicting VM performance to AWS. The methodology is confirmed using a cloud data file from Github and then improved by real -time rating with real data sets implemented on cloud AWS. This illustrates its efficiency in accurately predicting both applications and VM performance, with the finding underlining the pre -establishment of the GRU in cloud sources within the actual AWS contexts.