TL;DR: 👀 to https://maxday.github.io/lambda-perf/
to see the benchmark result:
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.
An ultra simple hello-world function has been written in each AWS supported runtime:
nodejs16.x
nodejs18.x
nodejs20.x
python3.8
python3.9
python3.10
python3.11
python3.12
dotnet6
java11
java11 + snapstart
java17
java17 + snapstart
java21
java21 + snapstart
ruby3.2
ruby3.3
in addition to the following custom runtimes:
go
onprovided.al2
go
onprovided.al2023
rust
onprovided.al2
rust
onprovided.al2023
c++
onprovided.al2
c++
onprovided.al2023
dotnet7 aot
onprovided.al2
dotnet8 aot
onprovided.al2
dotnet8 aot
onprovided.al2023
quarkus native
onprovided.al2
graalvm java17
onprovided.al2
graalvm java21
onprovided.al2023
apple swift 5.8
onprovided.al2
bun
onprovided.al2
(with and without layer)llrt
onprovided.al2023
Each of this function is packaged in a zip file, uploaded to a S3 bucket.
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.
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
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.
Hack/Fork/Send PR and create your own benchmarks!