Performance-by-Design: Embedding Intelligent Scaling and Guardrails Into Platform Engineering
Talkat Open Source Summit India 2026 · on Tuesday, 2026-06-16· at 14:00· Lotus Ballroom 3 (Level 3)·
summary
Claims a concrete method: codify SLOs, instrument app-aware metrics with Prometheus/OpenTelemetry, and enforce policies via OPA so autoscalers (HPA or KEDA) act on workload signals not just CPU. Reports a surprise: model load time = 100s breaks a 60s stabilization window, needed 120s and six pods instead of two. Shows reference stack, tenant isolation patterns, automated rollback/remediation, and measured gains: lower MTTR, better SLO compliance, and tighter cost control using performance-by-design, golden paths, and an observability pipeline.
AI-written summary · about
chapters (21)
- 0:00Introduction
- 1:28Why performance matters
- 4:17Performance by design explained
- 7:07Problems with traditional approaches
- 10:14Shift left performance
- 14:04Performance by design in platforms
- 15:31Design phase SLOs
- 17:19Automating testing and enforcement
- 18:47Golden paths and deployments
- 20:33Runtime scaling and control loops
- 24:03Reference architecture
- 26:27Intelligence and policy engine
- 28:14Remediation strategies
- 30:24Remediation actions and guardrails
- 32:11Open source implementation
- 33:39Policy examples for AI workloads
- 36:08Tenant isolation
- 37:55Multi-tenancy models
- 40:22Performance across isolation models
- 41:46Real-world impacts
- 43:53Conclusion
related talks
41mPerformance Engineering in the Age of DevOps and Agile
37mUsing SLOs for Continuous Performance Optimizations of Your K8s Workloads
36mO11y as Platform Contract: Instrumentation and CI/CD-Enforced SLOs
01h 02mTools for simplified large scale systems
54mUnraveling the Differences
31mServerless Observability: a case study of SLOs
39mPlatform Engineering as a (Community) Service
42mMaking Resilient, Self-Adaptive, And Autonomous Kubernetes Clusters