Scalability Intro – Shyam Jeedigunta, Google & Bob Wise, AWS (Any Skill Level)
summary
SIG Scalability runs CI scale tests mostly on GCE, executing 100–5,000 node tests with a density test (30 pods/node) and load tests that block releases. It uses Kubemark hollow nodes to simulate 5,000-node clusters on modest hardware, gathers CPU/memory profiles, and is building clusterloader for real-workload testing. Current SLOs include 99p pod startup <5s (no image pull) and 99p API call latency <1s; common failures are high churn, leaked Job objects, and API-server saturation.
AI-written summary · about
chapters (23)
- 0:00Introduction
- 0:50Agenda
- 1:26Scalability Update
- 3:01Community Meetings
- 3:30Google Scalability
- 6:33Performance Envelope
- 9:48Performance Grid
- 14:42Tips for Cluster
- 15:12Examples
- 17:58CueMark
- 19:37Scalability
- 23:33Future Work
- 25:45Scalability Framework
- 30:54Test Grid
- 31:46Pods
- 33:16Kubernetes throughput
- 34:07Kubernetes throughput for production
- 35:45Thinking to factors
- 36:49Cluster size
- 38:07Welldefined process
- 40:11Configurations
- 40:49Splitting
- 41:41Provider Dock
related talks
43mSystems @Scale Summer Q&A Week 2
30mBuilding distributed systems is hard - case study (...)
29mScaleCheck: A Single-Machine Approach for Discovering Scalability Bugs...
31mScaling a Distributed Container Management System: Diego Performance
38mScaling Kube-Apiserver to Thousands of Tenants – Lessons Learned From kcp
47mWeb-Scale IT Engineering.
46mPerfect Scalability
30mBuilding Stable, Scalable & Agile Distributed Systems at Enterprise Scale