Streaming Telemetry-Based Network Microburst Detection with Queue Metric Analytics for Packet Loss Mitigation

Network microbursts, defined as high-intensity traffic surges lasting for an extremely short duration 100–800µѕ, have become a major cause of hidden packet loss that degrades the performance of critical applications in modern data center infrastructure. Conventional monitoring methods based on SNMP (Simple Network Management Protocol) fail to detect this transient phenomenon due to inadequate polling intervals, creating a blind spot in network visibility. This research designs, implements, and evaluates a microburst detection framework that leverages streaming telemetry and multivariate analysis of queue metrics to overcome the limitations of existing systems. The study adopts the Design Science Research (DSR) approach in a Mininet emulation environment with Open vSwitch, utilizing an integrated pipeline of gNMI/gRPC, Prometheus, and a Python-based detection algorithm that combines dynamic thresholding and queue metric correlation analysis. Evaluation against 1.6 million microburst events revealed that the proposed framework achieved a detection accuracy of 96.8% with an equivalent F1-Score at a 10 ms sampling interval, dramatically outperforming SNMP, which failed to detect any events. Correlation analysis showed a strong relationship between queue depth and packet drop rate, confirming the effectiveness of queue metrics as predictive indicators. The multivariate algorithm successfully reduced the false positive rate by 63% (from 5.7% to 2.1%) compared to a static threshold approach, despite increasing CPU overhead by 8–19%. The results of the study demonstrate the effectiveness of streaming telemetry with queue metric analysis for real-time microburst detection, while also providing practical implementation guidelines in the form of optimal configurations at 10–30ms intervals for various deployment scenarios.

Keywords: Network microburst, Network monitoring, Packet loss, Queueing theory, Streaming telemetry