EXASAGE: The first data center operational data analysis assistant

Data centers increasingly rely on Operational Data Analytics (ODA) for real-time insights from vast streams of telemetry data. They typically utilize NoSQL databases for scalability and to handle diverse data types, which results in unstructured data representations and poses significant challenges for data retrieval and interoperability. Indeed, the lack of standardization, combined with schema flexibility and complex data structures, makes it difficult for system administrators to write and execute queries, ultimately complicating the automation of data retrieval tasks. Pre-trained Large Language Models (LLMs), with their latent knowledge, promise a ready-to-use AI-driven data interoperability layer, enabling data retrieval through natural language input. However, they often generate inaccurate or hallucinated query code when handling heterogeneous data sources and complex data structures. In this paper, we present EXASAGE, the first operational data analysis assistant for ODA, which leverages a Knowledge Graph (KG)-based approach to provide an AI-driven interoperable layer that addresses LLM limitations and simplifies data retrieval tasks in data center facilities through a prototype implementation. EXASAGE employs an LLM-based query generator to convert natural language into SPARQL queries (native to KGs), executed at a graph database endpoint, along with a virtual KG approach that dynamically generates KGs with only the data relevant to the user’s input query, significantly reducing the storage overhead associated with a fully materialized KG in such large-scale telemetry systems. In evaluations on 1000 user input queries, EXASAGE achieved a 93.6 % accuracy in generating correct SPARQL code and retrieving correct answers, significantly outperforming the 25 % accuracy of NoSQL/SQLite queries, which frequently exhibited severe hallucinations. Furthermore, SPARQL queries were generally more concise and demonstrate shorter inference and execution times compared to NoSQL/SQLite queries. The average end-to-end time for a single execution cycle was 12.77s, making it suitable for interactive, non-critical operational data analysis tasks. The maximum observed storage overhead across all generated virtual KGs was just 52.62 MiB.