Optimizing Istio Autoscaling: From Resource-Centric To Connection-Aware
Talkat Istio Day North America 2025 · on Monday, 2025-11-10· at 15:25· Building B | Level 4 | B406b-407·
summary
Operates Istio ingress scaling by replacing CPU/memory-only HPA with a downstream_cx_active connection metric plus a tuned curved metric (derived from empirical tests + AI) and a dual‑tier sizing policy for pods and nodes. Uses Istio ingress gateway (TLS passthrough, sidecars), Admiral/ctographer, AWS CLB/NLB constraints (1024 CLB queue, ~28k ephemeral ports), and HPA. Result: 67 pods (3 CPU) handle 0→216k TPS at <30% CPU, eliminate 503/504s and cut overprovisioning costs.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:45Scale at Intuit
- 3:10Service mesh overview
- 5:16Request flow layers
- 8:34Core scaling problems
- 10:17Resource-based limits
- 13:07Infrastructure challenges
- 16:20Solution objectives
- 19:32Hybrid autoscaling
- 21:38Dual-tier policy
- 22:43Results and improvements
- 23:46Cost optimization
- 24:49Q&A
related talks
40mThe efficient way of Autoscaling Your Workload in Kubernetes
18mLessons from Scaling AI Powered Translation Services Using Istio
47mIstio, le meilleur ami de votre cluster K8S
17mScaling Made Simple: Implementing Autoscaling for Optimal K8s Performance
36mIstio to the rescue
21mMusic Matched to Different Ways to Scale in K8s, What Could Go Wrong?
34mIstio Performance and Best Practices in Large Scale Kubernetes...
32mOptimizing Resource Usage in Kubernetes