How Ancestry Got Kubernetes to Run 2x Better Per Dollar Using AI
Talkat KubeCon + CloudNativeCon North America 2019 · on Wednesday, 2019-11-20· at 17:20· Room 15AB - San Diego Convention Center, Mezzanine Level·
summary
Ancestry hooks a machine learning optimizer into CI/CD and Kubernetes to run live canary experiments that tune pod-level settings — starting with CPU and memory then adding socket connections, thread counts, Java heap and GC — making about 40–50 changes/day to each canary. A PoC on the DNA‑matches microservice cuts resources by >50% with no performance drop, equivalent to 2x performance per dollar; broader rollout yields ~30% aggregate savings with some apps hitting 50–70%, saving hundreds of thousands to millions.
AI-written summary · about
chapters (7)
- 0:00Intro
- 0:30About Darek
- 6:04What does Ancestry do
- 11:40Infrastructure
- 12:21DNA Matches
- 15:09Moving Traffic
- 19:03Reducing Capacity
related talks
42mTaking the Helm - Ancestry's Journey to Kubernetes
33mUncover Kubernetes cost-optimization trends
40mHow an AI driven approach reduces cloud cost and makes your Kubernetes infrastructure autonomous
31mAccelerate AI adoption: automated cloud efficiency
18mLet the machines optimize the machines
40mOptimizing AI/ML Workloads
42mOptimize Kubernetes costs while matching your application SLO using AI
30mPortrait Service: AI-Driven PB-Scale Data Mining for Cost Optimization and Stability Enhancement