Automated Detection of Network Card Bottlenecks in Apache Pulsar: An Enhanced Framework with Dynamic Thresholds and Root Cause Analysis

Abstract

Apache Pulsar message queuing systems require robust hardware infrastructure monitoring to maintain optimal performance. Traditional static threshold-based alerting fails to account for hardware variations, workload characteristics, and transient issues, leading to false positives and delayed incident response. This paper presents an enhanced monitoring framework that addresses these limitations through an intelligent centralized analysis service that pre-computes optimal thresholds, detects configuration mismatches, performs root cause analysis, and emits actionable metrics. Unlike traditional architectures that place analysis burden in alert rules, our approach centralizes intelligence in a dedicated service, simplifying operations while improving accuracy. Validation across production environments demonstrates 96.3% detection accuracy, 1.4% false positive rate, 61% reduction in mean time to resolution (MTTR), and automatic generation of specific remediation recommendations.