Identifying Relevant Data Center Telemetry Using Change Point Detection
Daniel Alves, Katia Obraczka, Rick Lindberg · 2020
In this paper, we focus on the problem of data center performance monitoring, more specifically, how to manage the large volume of data generated by data center telemetry tools. We propose a framework that uses Change Point Detection (CPD) to identify sources of useful telemetry and based on that information, filters incoming telemetry data in real-time as the data center operates. To evaluate our proposed CPD-based telemetry triage framework, we conducted experiments using a small emulated data center driven by different workloads. We also report results from experiments with telemetry data collected from a privately-owned, commercial, multi-tenant data center. Preliminary experimental results show that our CPD-based tool can filter out significant amounts of irrelevant telemetry while preserving most relevant telemetry sources.