RT-LogAAS: A Real-time Log Anomaly Analysis System based on Large Language Models for Net-Cloud

Abstract

With the development of cloud computing technology, the infrastructure of telecom operators has been fully virtualized. Now it's running on Net-Cloud platforms more often. To provide the stability of Net-Cloud platforms, log analysis plays a crucial role. However, the logs of Net-Cloud platforms is not only formed by multi-sources, but also quite heterogeneous which bring great challenge to existing methods, especially when it comes to timeliness and effectiveness. To overcome these challenges, we propose RT-LogAAS, a real-time log anomaly analysis system based on large language models for Net-Cloud. By building a real-time log streaming engine and integrating LogLSHD with LogLLM technology, we try to enhance the accuracy and timeliness for log anomaly detection. Experimental results show that RT-LogAAS outperforms the current state-of-the-art solutions, especially in handling heterogeneous logs. Thus, RT-LogAAS is more suitable for application to industrial-grade log analysis tasks.