International Journal of Science, Engineering and TechnologyAuthors: Avinash Yeluri
Abstract: The reason why cloud-native enterprise applications form the backbone of today’s digital services is that they possess high levels of scalability, agility, and rapid deployability. However, because of the inherent distributed characteristics of microservices, containers, and cloud infrastructure, many reliability issues arise, including problems related to service unavailability, resource starvation, latency variations, and cascading failures. The traditional methods of monitoring these systems only include the use of reactive fault detection techniques and thresholding techniques, which are inadequate for predicting reliability problems in highly dynamic systems. In this approach, the integration of observability in cloud computing, reliability analysis, predictive learning, and adaptive risk assessment have been integrated into an architecture that can be used to predict the occurrence of reliability deterioration before any service interruptions occur. The proposed architecture uses machine learning techniques on telemetry data such as logs, traces, services performance metrics, resource utilization, and fault history information in order to produce reliability score and failure risk prediction. Experiments were done using large scale cloud native operations datasets and comparisons made to other models like DQN-SHVNF and LLM-DBN frameworks. According to the experiments, the proposed model has a Reliability Prediction Accuracy of 99.1%, Failure Risk Detection Rate of 98.6%, Service Availability of 99.5% and System Resilience Index of 99.4% which is much better than current models.
DOI: https://zenodo.org/records/22957505
