(N+1) Lessons I learned about serverless
summary
Practical serverless advice for AWS Lambda: measurement shows short Lambdas (~200 ms) scale to ~5,000 RPS per 1,000 concurrent containers but performance and RPS drop as execution time increases and providers may throttle around 1,000–1,500 concurrent executions. Recommendations include splitting monolithic functions into many small handlers, using SQS/SNS/EventBridge or Step Functions for orchestration, uploading files via S3 presigned URLs to reduce Lambda runtime, sharing code with Lambda Layers, enforcing least-privilege with IAM Access Analyzer and SAM templates, switching to Graviton CPUs for ~34% better cost/performance, and using Power Tuning and budget alarms to avoid runaway costs.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:14Disks
- 4:55Scaling the service
- 7:23Exposing the service to the user
- 8:51Exposing the service directly
- 11:28Reducing execution time
- 13:55Development permissions
- 16:52Improving performance and lowering costs
- 18:32Conclusions
related talks
15mAlternative Ways To Think About Serverless
46mServerless Architecture Patterns and Best Practices
43mServerless in Java Lessons learnt
21mBeyond the Basics: Production Serverless Patterns for Extreme Scale
35mEfficiency of Serverless Architectures with AWS Solutions
42mNo Server Is Better Than No Server: Building Blocks for Serverless Architectures
50mServerless Architecture Patterns and Best Practices
43mBest Practices for Serverless Developers