Modern information infrastructure comprises a wide range of products from
different vendors, making management and monitoring increasingly complex.
Vendor-specific proprietary tools make troubleshooting and problem resolution
difficult due to system heterogeneity, while centralised tools require complex
configuration and expertise in specific query languages. In addition, both
types of tools often involve costly licensing, and their advanced analytics are
performed in the cloud, which may not be acceptable. To address these challenges, we implemented an internal infrastructure monitoring system based
on generative artificial intelligence. The system identifies relevant sections
within large volumes of unstructured system logs, semantically correlates
information from different sources, and generates structured reports. This
enhances the efficiency of system engineers while eliminating concerns regarding sensitive data, as all processing is performed within the internal network.
We conducted analyses using two different graphics cards and various large
language models, selected configurations appropriate for our requirements,
and performed cost assessments.
|
