Rust is 9.8x faster than Node.js.
I agree, Rust is not the simplest programming language.
C# is 7.5x faster than Node.js, and .NET is very accessible.
I ran all these tests after seeing so many job offers claiming "building a top team to deliver super performant APIs with Python or Node.js".
Please stop that. This is not true. Saying that is like running a decathlon with stones in your shoes.
I also spent some time comparing different JVMs, both in Docker and without Docker.
Comparing Java Native vs Java in Docker => You lose a lot of performance using Docker.
Be smart, choose the right stack, and please stop using Python and Node.js for APIs!
You will save time, achieve better performance, and keep your team happy.
If you need Python for AI, then go for https://aspire.dev. C# backbone with Python workers.
Machine: OVHcloud VPS-2, 6 vCore, 12 GB RAM, unlimited traffic for 7€/month tax included.
Discution on LinkedIn
In this repository you will find simple implementations of a UserToken API in multiple programming languages.
The objective is to get the fastest response for each technology.
We can get better performance by optimizing the code, or choosing the right runtime.
For Java minimal API is faster than Spring Boot.
For Typescript or Javascript, Bun is faster than Node.js. Bun sqlite implementation is really optimized.
sequenceDiagram
participant Client
participant API
participant SQLite
Client->>API: POST /api/auth/get-user-token
Note over Client,API: JSON: {"UserName": "[email protected]", "HashedPassword": "sha256hash"}
API->>SQLite: SELECT id FROM users WHERE mail = ? AND password = ?
SQLite-->>API: Return user_id or null
alt User found
API-->>Client: 200 OK {"user_id": 12345}
else User not found
API-->>Client: 401 Unauthorized {"message": "Invalid credentials"}
end
Note over Client,API: Special case: UserName="no_db" returns user_id=12345 without DB query
A test client is provided in C# .NET 9 in directory dotnet_test. Look at readme.md in this directory for instructions.
To be sure api calls are real I added some trace in the console that display userId to be sure we get a real userId from the database.
This project is open to any improvement, please fork and create a pull request.
You should keep the same API, same database (sqlite), same data (10000 users) and same test procedure (100000 requests, 16 concurrent clients).
Please test on your machine the current version and compare your improvement. Please commmit only if you get really better performance.
Result below is from my machine under Windows 10, AMD Ryzen 7 2700X, 32GB RAM, SSD NVMe.
I keep the best result after 5 runs.
On Windows 10
The Debian tests are the most interresting because we do not put production on Windows 10.
We can observe a real difference between the two platforms. Results are more expected on Debian.
I didn't tried C++ on Debian because C++ is too complex today to start a new project and it's very tricky and time consuming to obtain good results.
This project is for educational and benchmarking purposes.
Client and server are on the same machine, so network latency is not taken into account. Also when a server is very fast, the client use half of the CPU to send requests. This is an unfair advantage for slow servers.
20251102 - SebDeLille - Proposal to improve rust-actix, gain 5700 req/s
20251102 - Kuurama - C# class to record to gain 1000 req/s
