Q&A Session 1 – Neil Spring
summary
Q&A covers concrete results and operational tools: Millisampler shows short, frequent microbursts that cause loss when contention is high, with contention varying substantially even within one or two seconds, so buffer sharing and congestion control must be tuned per rack and adjusted dynamically; Millisampler wakes every few minutes, uses sketches so memory is low, and adds only millisecond-to-microsecond CPU latency. Transport-layer synthetic blocks measure delivery without app changes and apply inside the data center; SLO derivation is product-specific, continuously monitored and rederived when predictive power drops, and there is automated detection of synchronized bursts for alerting.
AI-written summary · about
chapters (12)
- 0:00Welcome
- 0:46Microburst analysis insights
- 3:36Measuring network conditions
- 5:03Service-level objectives
- 6:30Network monitoring costs
- 8:17Network characterization approaches
- 10:21Evolving SLOs over time
- 12:34Delay-based congestion control
- 14:45Network characterization for health
- 16:30SLO compliance patterns
- 19:25Detecting app synchronization
- 21:11Closing
