查看文章

Short-lived bursts of traffic, known as microbursts, can lead to packet drops, jitter, and performance degradation potentially impacting data transfers. Detecting microburst events on production networks is challenging due to the sub-second monitoring interval required and the amount of data produced. We propose an adaptive and efficient approach to detect and store microbursts leveraging In-band Network Telemetry (INT) and an enhanced telemetry collector capable of adapting its routines to the bandwidth utilization. Field trials demonstrated our capabilities to detect millisecond-based microbursts and an efficient persistence data gathering approach with data reduction above 60%, while preserving good levels of accuracy. Source code and measurement datasets are public.