How AIOps Can Turn IT Operations into Fully Unmanned Systems

Professor Pei Dan from Tsinghua University presents a comprehensive view on "Unmanned Operations based on AIOps," highlighting the growing mismatch between human‑driven decision making and the complexity of modern distributed systems.

Unmanned Operations Rating

The proposed rating uses the metric Cores per Op (CPO), i.e., the number of X86 CPU cores managed per weekly 40‑hour operation (Op). Higher CPO indicates a higher level of automation. Example calculations show Level 0 for traditional industries, Level 1–3 for internet companies, and illustrate both horizontal (cross‑company) and vertical (time‑based) comparisons.

Achieving Unmanned Operations with AIOps

AIOps serves as the "brain" that consumes monitoring data (the "eyes") and drives automated actions (the "hands"). It also builds an operations knowledge graph for query and decision support. The architecture separates three modules: perception, decision, and execution, each requiring tailored AI algorithms.

Intelligent Fault Detection

The lab’s research on single‑metric anomaly detection evolved from supervised methods (IMC 2015) to unsupervised approaches (WWW 2018) and further enhancements such as temporal modeling, GAN‑based noise handling, clustering, transfer learning, and semi‑supervised techniques for short‑lifecycle services. These algorithms have been integrated into a unified "Intelligent Fault Discovery" system that automatically pinpoints fault locations, suggests remediation, and correlates with historical incidents.

Operations Knowledge Graph

The knowledge graph automatically extracts entities (servers, containers, services, metrics, etc.), their attributes, and relationships from operational data, providing a centralized, queryable, and continuously updated view that surpasses traditional CMDB and expert‑knowledge approaches. Examples illustrate hardware‑software mappings, dynamic metric profiles, and fault‑propagation graphs that enable precise root‑cause analysis and capacity planning.

Conclusion and Outlook

The speaker emphasizes that while the journey to fully unmanned operations is demanding, the combination of quantitative rating, AIOps architecture, intelligent fault detection, and knowledge‑graph techniques provides a solid foundation. Ongoing research and community collaboration are essential to realize the transformative impact of AIOps across industries.

This article is compiled from Professor Pei Dan’s latest speech at the 10th GOPS Shanghai session.