Task failure prediction in cloud computing systems

As cloud data centers grow in scale and complexity, ensuring high service reliability and minimizing failures have become critical challenges. Despite technological advances, failures due to hardware and software issues persist, disrupting tasks, wasting resources, and impacting service reliability. Accurately predicting task or job failures before they occur is essential to reducing downtime and unnecessary resource usage. Traditional fault-tolerance methods like checkpointing and replication are insufficient for the complexity of modern systems. Consequently, machine learning and deep learning techniques have been adopted to analyze system logs and predict failures more accurately. Federated learning further enhances this by enabling decentralized data analysis across nodes, preserving privacy while improving prediction accuracy through collaborative learning. In this paper, we propose a fault prediction mechanism based on federated learning and a deep neural network to identify patterns leading to task failures. Our model achieved a high prediction accuracy of ۹۵.۳%, making it a robust solution for failure prediction in cloud computing environments.

Keywords:

federated learning , cloud computing , multi-layer perceptron , fault tolerance