🔄 Continous Lambda Cold Starts Benchmark

TL;DR: 👀 to https://maxday.github.io/lambda-perf/

to see the benchmark result:

Why?

There are already a lot of blog posts talking about Lambda Cold Starts performance per runtime but I could not find any always up-to-date information.

That's why I decided to create this project: data is always up to date as the benchmark is running daily.

How does it work?

Architecture diagram

Step 1

An ultra simple hello-world function has been written in each AWS supported runtime:

• nodejs14.x
• nodejs16.x
• nodejs18.x
• nodejs20.x
• python3.7
• python3.8
• python3.9
• python3.10
• python3.11
• dotnet6
• go1.x
• java8
• java8.al2
• java11
• java11 + snapstart
• java17
• java17 + snapstart
• ruby2.7
• ruby3.2

in addition to the following custom runtimes:

• go on provided.al2
• go on provided.al2023
• rust on provided.al2
• rust on provided.al2023
• c++ on provided.al2
• c++ on provided.al2023
• dotnet7 aot on provided.al2
• quarkus native on provided.al2
• graalvm java17 on provided.al2
• apple swift 5.8 on provided.al2
• bun on provided.al2 (with and without layer)

Each of this function is packaged in a zip file, uploaded to a S3 bucket.

Step 2

Every day, each function is freshly grabbed from S3, deployed and invoked 10 times as cold starts.

Then the REPORT log line containing the init duration, max memory used and other useful information is saved to a DynamoDB table.

Step 3

After all these invocations, all information stored in DynamoDB is aggregated and a new JSON file is created, then commited to this repo. ie: https://github.com/maxday/lambda-perf/blob/main/data/2022-09-05.json

Step 4

A static website, hosted on GitHub pages here: https://maxday.github.io/lambda-perf/ fetches this JSON file and displays the result in a (nice?) UI.

Step 5

Hack/Fork/Send PR and create your own benchmarks!

Disclaimer

⚠️ This project is not associated, affiliated, endorsed, or sponsored by any companies nor have they been reviewed tested or certified by any company.