Go FaaSter: Cold Start Optimization in a Serverless Platform
summary
Describes Tencent Serverless Cloud Functions (SCF) and the specific causes of cold-start latency across networking, infrastructure, and deployment. Explains architectural changes—elastic network interface improvements, moving from containers to microVMs, faster function code deployment, and resource reuse—and how combining machine learning with autoscaling can prevent cold starts entirely.
AI-written summary · about
chapters (17)
- 0:00Intro
- 1:571.1 Tencent Cloud SCF
- 2:401.2 Challenges we meet
- 3:471.3 SCF Architecture
- 6:542.1 What is function cold start?
- 8:562.2 Function cold start stages
- 11:463. Several ways to optimize cold start
- 13:113.1.2 How does it work?
- 18:023.2 Code Caching
- 19:463.3.1 Auto Scaling
- 25:163.3.3 Predictable Prewarming
- 27:413.4 Function Network Model
- 28:463.4.1 Legacy Network - access VPCB
- 31:023.4.2 SCF Solution - access VPC
- 32:463.4.3 Legacy Network - access Internet
- 34:243.4.4 SCF Solution - access Internet
- 35:07Conclusion
related talks
37mGo FaaSter: Cold Start Optimization in a Serverless Platform
27mServerless Performance on a Budget
33mGoing FaaSter: Cost-Performance Optimizations of Serverless on Kubernetes
25mCF Serverless: CloudFoundry as a Platform for Serverless Computing
37mState of Serverless
58mServerless Under the Hood
51mFaaS or not to FaaS? Visible and invisible benefits of the Serverless paradigma
12mFaaSNet: Scalable and Fast Provisioning of Custom Serverless Container Runtimes...