Skip to content

Latest commit

 

History

History
94 lines (65 loc) · 3.61 KB

readme.dev.md

File metadata and controls

94 lines (65 loc) · 3.61 KB

Rust framework benchmarks

Benchmarking utility to test the performance of all the rust web frameworks. Built with rust 🚀.

Demo

Demo

(Last updated: Wed Jul 27 2022 02:59:45)

Frameworks included

Nickel - An expressjs inspired framework for rust

Results

Concurrency: 2048 Duration: 45 secs Threads: 2
Name Req/sec Avg Latency Min Latency Max Latency Std Dev 95% 99% 99.9% # Requests Transfer Rate # Errors
Nickel 20,128 96.23ms 14.54ms 1340.39ms 147.12ms 701.37ms 945.37ms 1061.11ms 905,718 39.36KB/Sec 894756

Benchmarking tool

The benchmarks have been performed using rewrk, locally.

Check the raw output from rewrk here.

Try it yourself

Everything is automated, including adding a framework, generating md file output, and running the tests without having to start all the servers at once!

Pleas make sure you do not have capped soft limit or hard limit for file descriptors, this may cause benchmarks with high concurrency (-c) fail with OS error 54, to fix that -

Linux

  1. Open /etc/sysctl.conf
  2. Add fs.file-max = 65536
  3. Reboot your pc and verify it after rebooting with sysctl -p

Mac

  1. Check your current limit - launchctl limit maxfiles
  2. The first entry is the soft limit, and the second entry is hard limit. In most cases the hard limit is unlimited, it's the soft limit that we need to change
  3. Update the hard limit sudo launchctl limit maxfiles 65536 200000
  4. Reboot.

Alternatively, if you wish to change the soft limit to only run the benchmark one time, you can change it for your current terminal session by ulimit -n 65536. This doesn't require boot, and will reset back to the default (usually 2560) after restarting the terminal.

To run the tests locally, please follow the steps -

  1. Download the repository as a zip, or clone/fork it.
  2. cd rust-framework-benchmarks
  3. cargo build --release --bins
  4. Run the main script and you're good to go.. ./target/release/main or cargo run --release --bin main

All the output will be stored in perf/{name}/{concurrency}.txt*

Machine used

M1 Max MacBook Pro 2021 - 64GB ram, 10 CPU cores and 32 GPU cores

Suggestions and changes

All the suggestions, code changes or additions of another web framework is appreciated. I'd like to keep the code as close as a real world scenario, instead of optimising it to the metal.

To add a new library/framework, please make sure to use the PORT provided through the benchmark dynamically. Default is 3000 for all. You can change it in config.json.

Also, to add a framework, add an entry inside config.json for the benchmarks to detect it.

[
  {
    // Name of your framework. Displayed in the readme and during logs
    "name": "Axum", 
    // Default for all the frameworks
    "port": 3000,
    // the name of the file that your framework is located
    "binary": "axum", 
    // Github or Crates.io or Website
    "url": "https://github.com/tokio-rs/axum" 
  }
]
fn get_port_number() -> String {
    std::env::args().collect::<Vec<String>>()[1].clone()
}
 Server::build().bind("hello-world", format!("127.0.0.1:{}", port_number))