Going FaaSter: Cost-Performance Optimizations of Serverless on Kubernetes

on YouTube ↗33m

" summary

Walks through concrete runtime fixes for cold starts using Fission on Kubernetes and general FaaS models, showing four actionable optimizations: instance reuse, stem-cell pooling to avoid runtime startup, artifact prefetching, and predictive pre-warming. Compares tradeoffs like concurrency=1 vs 50 (more isolation yields more cold starts), public clouds keeping containers alive for hours to days versus Kubernetes systems that tear them down in minutes, and notes pooling creates a visible fixed cost rather than per-invocation billing.

AI-written summary · about

" chapters (4)

1. 0:00Start
2. 10:45Function Resource Reuse
3. 15:16Optimization 2 Function Runtime Pooling
4. 18:40Optimization 3 Function Prefetching

" related talks

• 37mGo FaaSter: Cold Start Optimization in a Serverless PlatformScott Zhou, Yanbo Li · KubeCon + CloudNativeCon + Open Source Summit China 2019
• 37mGo FaaSter: Cold Start Optimization in a Serverless PlatformScott Zhou, Yanbo Li · KubeCon + CloudNativeCon China 2019
• 27mServerless Performance on a BudgetErwin van Eyk · KubeCon + CloudNativeCon China 2018
• 39mFaaS: A Contemporary, Comparative StudyDilleswara Anupoju · Cloud Foundry Summit North America 2019
• 33mFission: Serverless Functions for KubernetesSoam Vasani · KubeCon + CloudNativeCon North America 2017
• 36mFission: a Serverless Function Framework on Kubernetes [B]Soam Vasani · CloudNativeCon + KubeCon Europe 2017
• 34mServerless Operations: From Dev to ProductionErwin van Eyk · KubeCon + CloudNativeCon Europe 2019
• 21mWhat is the Future of Serverless Computing?Michael Maximilien · KnativeCon North America 2022