How to be data-driven when you aren’t Netflix (or even if you are)
summary
Built an analytics pipeline that scales with a tiny team, using Kafka and Druid and Flink plus Hadoop, Zeppelin and PySpark to collect and explore metrics; traffic randomization, canary/A/A′ deployments and configuration injection provide experiments at user level; histograms, chi‑square tests and bucketed visualizations quantify significance and locate effects; result claims include 20 million unique sessions a day and a reduced buffering by 20% outcome on ABR tweaks, with nuanced splits showing +5% offload for VOD and 3% for live.
AI-written summary · about
chapters (17)
- 0:00Intro
- 3:01The Challenge: Axioms - Requirements
- 4:27The Challenge: use cases
- 6:32How it Started: Our Basic Data Lab
- 8:43Potential Upgrades Based on Your Needs
- 10:27AB Testing: Deployment
- 11:05AB Testing: Configuration Injection
- 11:41AB Testing: New version roll-out
- 14:30101: Outliers
- 15:19102: Simplon & Confounding Factors
- 17:52103: Histograms
- 18:43201: Tackling noise
- 20:08201: Statistical Significance
- 21:07201: Chi-Squared Test
- 22:24201: Chi-Squared Visualized
- 23:52201: Population movements: practical significance
- 25:25201: Impact on statistical significance
related talks
21mPutting big data self-service analytics in the hands of business people
07mTransforming Big Data into Real-Time Product Insights
21mData-driven approaches in a technology startup
50mData-Driven Engineering
45mUsing time-series data to better understand our products & infrastructure
41mFast Data - an Overview of the Concepts, Architecture and Technology of Streaming Analytics
29mData Acquisition: Overcoming Challenges
27mTales at Scale: Analytics at 1000 QPS and Beyond