On-demand Container Loading in AWS Lambda
summary
AWS Lambda increases deployable artifacts from zip files to 10 GiB images while keeping cold starts low by chunking flattened container tar layers, content-addressing chunks in S3, and doing block-level demand loading into microVMs. The system uses consistent hashing, convergent encryption for deduplicable-but-confidential chunks, and erasure-coded caching (5-of-4 erasure coding) across tiers (L1 local ~65% hits, L2 datacenter, L3 S3 <1% misses). Result: ~40× package size increase, support for 15,000 containers/sec per customer, millions of requests/sec, and hundreds of trillions of invocations for >1M customers.
AI-written summary · about
chapters (9)
- 0:00The Lambda evolution
- 2:40Motivation for change
- 5:13Architecture and design
- 6:59Worker node implementation
- 8:48Efficiency and deduplication
- 11:19Security and encryption
- 14:28Caching and reliability
- 17:42Performance metrics
- 18:55Future and summary
related talks
35mContainers and AWS Lambda: Choosing optimal compute for modern apps
54mContainerless in the cloud with AWS Lambda
21mThe best compute for your storage - Amazon S3 & AWS Lambda
39mHow to Go with AWS Lambda?
46mHighly Scalable Image Storage with AWS Serverless
33mContainers and AWS Lambda: Choose the optimal compute for applications
47mReal Use Cases for Using AWS Lambda for Infrastructure Tasks
50mDeveloping Highly Scalable Image Storage Solution with AWS Serverless