NetPanel: Traffic Measurement of Exchange Online Service
summary
NetPanel measures Exchange Online traffic at component granularity by fusing IPFIX, ETW, application logs and applying IP→location translation, port aggregation, feature splitting, and top‑pair selection (top 1,000 machine pairs) to cut petabytes to gigabytes. It runs in production over a year, costs <1% CPU on servers and <0.01% of EXO cores for analytics, reduces stored data by tens of thousands×, and uncovers actionable issues (misrouted endpoint, legacy sender, compression bug) that cut long‑haul traffic 20% and help save tens of millions per year under a top‑heavy distribution of traffic.
AI-written summary · about
chapters (8)
- 0:00Start
- 3:36NetPanel Design Considerations
- 5:49NetPanel Overview
- 7:02Data Processing
- 9:20Production Insights
- 10:39Case Study 1
- 13:16Evaluation
- 14:26Conclusion
related talks
29mFinding the Needle in a Haystack: Tenant-Level Network...
08mNetwork Performance Monitoring
30mPanel: Telemetry That Matters
26mApplication performance management w/ open...
15mBuilding a Network Telemetry Platform
19mNetwork X-Ray Vision: eBPF for Cloud-Native Observability
26mUnified End-to-End Observability: How Comcast Generates SpanMetrics at Enterp...
34mNetwork Observability with eBPF and OpenTelemetry