Overview
Explore cold start optimization techniques in serverless platforms through this conference talk by Scott Zhou and Yanbo Li from Tencent. Dive into the architecture of Tencent Serverless Cloud Functions (SCF), examining factors contributing to cold start latency. Learn about architectural improvements, including elastic network interfaces, micro VM migration, function code deployment, and resource reuse. Discover how combining machine learning with autoscaling can potentially eliminate cold starts altogether. Gain insights into SCF's network models for accessing VPCs and the internet, and understand the challenges faced in serverless environments.
Syllabus
Intro
1.1 Tencent Cloud SCF
1.2 Challenges we meet
1.3 SCF Architecture
2.1 What is function cold start?
2.2 Function cold start stages
3. Several ways to optimize cold start
3.1 Micro VM System
3.1.2 How does it work?
3.2 Code Caching
3.3.1 Auto Scaling
3.3.3 Predictable Prewarming
3.4 Function Network Model
3.4.1 Legacy Network - access VPCB
3.4.2 SCF Solution - access VPC
3.4.3 Legacy Network - access Internet
3.4.4 SCF Solution - access Internet
Conclusion
Taught by
Linux Foundation