CoreSec Improves Datacenter Network Root Cause Analysis

Identifying the root cause of issues in large datacenter networks, particularly those built on Clos fabrics, is notoriously difficult due to noisy, partial, and asynchronous telemetry data. Traditional score-based root cause analysis (RCA) methods often fail under these conditions, leading to unstable or incorrect problem attributions. Researchers have developed CoreSec, a production-grade RCA system that addresses these challenges by employing an "abstention algebra" rather than relying on weighted fusion. This system uses control flags to enable telemetry agents to make deterministic decisions, and crucially, to explicitly abstain when the available evidence is ambiguous. This structured composition, combined with topology-aware configurations that map failure surfaces across Clos fabrics, ensures that CoreSec's conclusions converge monotonically as more evidence becomes available. Deployed at hyperscale, CoreSec has demonstrated stable and explainable RCA behavior across diverse network environments without requiring constant retuning. This approach provides a practical and robust foundation for automated root cause analysis in complex, real-world cloud networks, improving operational efficiency and reliability.