Systems @Scale Summer Live Q&A Session #2
summary
Hardware monitoring optimizes availability by balancing measurable cost against detected benefits with a simple cost/benefit equation, and anomaly detection runs on out-of-band telemetry with feature engineering plus tradeoffs between fleet-wide and cohort-specific ML models. A new container image system called Daddy deployed across Alibaba accelerates all container workloads and is being open-sourced and integrated into elastic block service. Multi-product regressions get handled by machine learning correlation pauses tasks that surface likely root causes within a few ranked candidates.
AI-written summary · about
chapters (12)
- 0:00Systems at Scale Q&A intro
- 0:47Hardware monitoring costs at NASA
- 2:31Container images at Alibaba
- 3:57Multi-product failures with ML
- 7:04Legacy integration at Google
- 8:30Istio rate limiting
- 8:51Federation vs agile teams
- 10:59Hardware failure analysis
- 12:00Open source at scale
- 15:34Anomaly detection at NASA
- 17:49Reliability engineering
- 18:53Closing remarks
related talks
38mData Secrets from a Platform Engineer
14mHere, There and Everywhere: Spotting Data Opportunities 101
33mMonitoring Systems as a Service – Walking the Line between...
11mScaling MLOps with Self-Service Platforms
52mWeek 3 Live Q&A
26mWeek 1 Live Q&A
32mMaking Models Work: How an optimized ML lifecycle can drive better ROI
26mML Product Experiments at Scale