Running Services at Google Scale
summary
Google claims its stack lets services scale from zero to billions without rewrites, running 4 billion containers/week on Borg-inspired orchestration and GKE with four-way autoscaling (pods, clusters, right-sizing). Kubernetes scale rises to 5,000-node clusters with bin-packing work and customers running up to 40,000 nodes; Pokemon Go exposed a 50× surge that drove fast autoscale features. Anthos bundles GKE and SGO/sto for site reliability, and eBay gains global service visibility and reduced ops complexity.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 0:13When to scale
- 0:47Problems with scaling
- 1:20Tips and tricks
- 2:06Whats the secret
- 2:28Google Cloud
- 3:40Kubernetes
- 4:58GKE
- 5:13Scaling Kubernetes
- 8:17Observability
- 9:09eBay example
- 10:41Anthos
- 11:14Switching over
- 11:24Service Mesh
- 13:57Cultural Change
- 15:39Summary
related talks
39mHow Google Cloud Hosts and Manages Kubernetes at Scale
42m2000 Nodes and Beyond: How We Scaled Kubernetes to 60,000-Container Clusters
07mKeynote: High Performance Computing on Google Kubernetes Engine
43mCloud Native - effective computing at scale
06mKeynote: Open Innovation: Kubernetes on Google Cloud
58mKeynote: So, I have all these Docker containers, now what?
41mKubernetes in Action: Examples from Google Container Engine
38mScaling Organizations Using Kubernetes