Cloud FinOps & Kubernetes Optimisation at Scale
summary
Expedia Group processes billions of cloud billing line items into a Spark/Hive S3 data lake, reconciles missing instance types, and aggregates to application as the primary key for thousands of apps and hundreds of clusters. Built a nightly cwatch recommendations engine and a chargeback model using PCU unit-rate chargebacks to bill tenants, running across 400+ EKS clusters, ~200k pods and ~12k instances. Found that at large scale building in-house (Redshift/Elastic/Kibana, Backstage) can be cheaper than vendors, and added developer-facing tooling like optimade PR comments to surface cost impact before merge.
AI-written summary · about
chapters (10)
- 0:00Intro
- 0:49What is FinOps?
- 6:00FinOps basics
- 10:24Expedia Group's FinOps journey
- 18:14FinOps platform chargebacks
- 20:40Kubernetes optimization
- 26:10How to measure effectiveness in FinOps negotiations?
- 32:17Optimization tradeoffs
- 47:20Summary
- 48:14Outro
related talks
27mScalable by Design: How to Build an Elastic Cloud Infrastructure...
31mKeeping Your Cloud Footprint in Check
36mFinOps tips for optimizing your cloud infrastructure and data costs
31mThe AWS Billing Machine and Optimizing Cloud Costs
27mUsage based cost optimisation at scale
47mCost optimization
38mFinancial Resiliency Engineering: Taming Cloud Costs
14mAccelerating Innovation at Expedia with OpenStack