Adaptive performance monitoring in cloud environments via recurrent neural networks

The study aimed to develop an adaptive methodology for analysing the performance of cloud computing infrastructures to improve the efficiency of resource management and reduce maintenance costs. The research addressed the implementation of the latest approaches to automate monitoring and analysis processes. The research methodology included the integration of data from monitoring platforms (Amazon Web Services CloudWatch, Google Cloud Monitoring, Prometheus) to collect key performance indicators. Data processing was conducted using Python libraries (NumPy, pandas, scikit-learn) to detect anomalies and generate time series. Recurrent neural networks and long-short-term memories based on TensorFlow and PyTorch were used to model performance. The implementation of continuous learning was used to adapt the models to the changing conditions of cloud systems in real-time. The main results of the study include the creation of an innovative system for predicting key performance metrics of cloud infrastructures with high accuracy. This was confirmed using the mean absolute error and root mean square error metrics. Real-time data integration was provided through the Amazon Kinesis platform, and visualisation and management were performed using Amazon CloudWatch and Grafana dashboards. Virtual machines and containers interacted with Nova, Glance, Cinder, and Neutron modules, and the Keystone module provided security through authentication and authorisation. Automatic resource scaling based on neural networks optimised the use of computing, network and storage resources. The developed methodology can be used to automate the management of cloud resources, reducing the need for manual intervention and cutting costs. The proposed method provided high speed due to interaction via REST and HTTPS and collected data in a time series format for primary processing. The integration of OpenStack with Apache Spark and the use of a high-speed data channel has increased the efficiency of the infrastructure. The findings demonstrated that the implementation of this methodology significantly increases the efficiency of cloud infrastructure management