Title

AWS re:Invent 2022 - Optimizing performance for .NET on AWS Lambda (XNT403)

Summary

Speakers: Saikat Banerjee (Engineering Manager at AWS) and Bo Goss (Senior Software Development Engineer at AWS).
Focus: Performance optimization for .NET applications on AWS Lambda, specifically addressing cold starts and introducing native Ahead-of-Time (AOT) compilation.
Key Points:
- AWS launched Snap Start for Java to improve cold starts and is looking to do the same for .NET.
- Cold starts are a significant issue for compiled languages like .NET and Java due to the initialization phase.
- Strategies to prevent cold starts include scheduled pings and provisioned concurrency, but these can be unreliable or costly.
- Code optimization and using smaller functions can help reduce cold start durations.
- Native AOT compilation is a focus for AWS, with tooling launched to support it on Lambda.
- Native AOT offers significant performance improvements over Just-In-Time (JIT) compilation, especially for cold starts.
- The session included a demo of native AOT vs. JIT compiled Lambda and a discussion on future improvements.

Performance and Cost: Optimizing Lambda functions is not only about improving end-user experience but also about reducing costs. More resources can lead to higher costs, but if the invocation time is reduced, overall costs can be lowered.
Cold Starts: The initialization phase of a Lambda function is where cold starts occur, taking more time to launch a .NET or Java application. Native AOT can reduce this time by front-loading the compilation process.
Native AOT: This is a significant investment area for AWS, with the potential to change the game for .NET on Lambda. It compiles code ahead of time, reducing the need for JIT compilation at runtime and thus improving cold start times.
Deployment: Deploying native AOT requires building in a Docker container to match the Amazon Linux 2 environment of AWS Lambda. This process can be resource-intensive and may require scaling up build resources.
Trimming: Native AOT always trims unused code, which can lead to smaller package sizes and lower memory usage. However, developers need to be cautious about runtime errors due to trimmed code that might be needed.
Future of .NET on AWS: Native AOT is seen as an important part of the future of .NET on AWS, with ongoing support and improvements expected. AWS encourages feedback from developers who try out these performance improvements.