The Big Kubernetes Mistake Paying More Than Necessary
summary
Kubernetes clusters often overprovision resources: a survey across more than 15 companies found about 10% CPU utilization and 23% memory utilization. The talk describes an AI-driven system that automatically resizes workloads, rebalances instances using beam-packing, diversifies spot instances, and improves autoscaling, claiming 50–80% cost reductions (one case 67%). It also predicts spot interruptions to extend warning lead time and performs live pod migration with roughly 2s interruption, and is running in production.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:09Overprovisioning problem
- 5:21Resource waste alarming
- 7:29Manual work risks
- 9:39Suballocation risks
- 12:09Cost inefficiency causes
- 13:53Automated optimization
- 17:10Shift right paradigm
- 20:25Smart pod packing
- 23:47Spot instance challenges
- 27:25Live migration disruption
- 30:19Cost savings results
- 32:43Call to action
related talks
33mUncover Kubernetes cost-optimization trends
26mDescubra las tendencias de optimización de costos en Kubernetes
29mRight-Size Your Kubernetes Deployments
44mFinOps at scale: Lessons from scaling Kubernetes workloads
21mLet your Kubernetes environment flourish
31mFailing to Autoscale?
47mOptimizing Kubernetes Resource Requests/Limits for Cost-Efficiency and Latency
31mAccelerate AI adoption: automated cloud efficiency