4 ms·

Has anyone compared Crystal and Rust on similar lines?

unixhero 6y ago
It would be cool to see. Crystal-lang isn't finished yet tho. Maybe it's a little too early for that?
tracker1 6y ago
Did a rust build with the following... fn fibonacci(n: u32) -> u32 { match n { 0 => 1, 1 => 1, _ => fibonacci(n - 1) + fibonacci(n - 2), } } fn main() { println!("{}", fibonacci(47)); } size: 2.67kb time: 7.391s I think rust wins :-D --- edit: not sure if I used valgrind/massif right, but the largest values in the output file. mem_heap_B=1816 mem_heap_extra_B=32 I think this means it stayed under 2k ram?
- tracker1 6y ago
  For the web server... 3.6mb file using hyper, not the leanest but one of the more popular options. extern crate hyper; use hyper::Server; use hyper::server::Request; use hyper::server::Response; fn hello(req: Request, res: Response) { let response = format!("Hello from {}!", req.remote_addr.ip()); res.send(response.as_bytes()).unwrap(); } fn main() { Server::http("127.0.0.1:3000").unwrap() .handle(hello).unwrap(); } Output from wrk -t8 -c400 -d60s http://localhost:3000/ http://localhost:3000/ Running 1m test @ http://localhost:3000/ 8 threads and 400 connections Thread Stats Avg Stdev Max +/- Stdev Latency 173.31us 88.10us 0.98ms 65.18% Req/Sec 35.06k 13.46k 45.48k 74.99% 8383194 requests in 1.00m, 775.50MB read Socket errors: connect 0, read 0, write 229, timeout 0 Requests/sec: 139487.00 Transfer/sec: 12.90MB According to top, rust_http is using about 2.8mb ram, 630% cpu, and wrk is using 700k and 370% cpu. Though I'm on completely different hardware, but the thing definitely screams... For reference: AMD Ryzen R9 3950X, 64GB DDR4@3200(CL18) under WSL2-Ubuntu (Windows 10 insiders)
- sk0g 6y ago
  Wouldn't processing time depend on your CPU, and unless you run the benchmarks in the article, this comparison would be meaningless?
  - tracker1 6y ago
    Don't have a go environment setup, but could do that. The executable size and memory use was also lower.
    - sk0g 6y ago
      Well to be honest, Go binary sizes are always going to be big, to the benefit of ease of development and deployment. Memory usage, I don't know. The Fibonacci benchmark code was a bit... shit. Do you want to send over your Rust code in a Gist or something, and I can compile and compare?
      - tracker1 6y ago
        It's in the comment above...

It would be cool to see. Crystal-lang isn't finished yet tho. Maybe it's a little too early for that?

Did a rust build with the following... fn fibonacci(n: u32) -> u32 { match n { 0 => 1, 1 => 1, _ => fibonacci(n - 1) + fibonacci(n - 2), } } fn main() { println!("{}", fibonacci(47)); } size: 2.67kb time: 7.391s I think rust wins :-D --- edit: not sure if I used valgrind/massif right, but the largest values in the output file. mem_heap_B=1816 mem_heap_extra_B=32 I think this means it stayed under 2k ram?

tracker1 6y ago
For the web server... 3.6mb file using hyper, not the leanest but one of the more popular options. extern crate hyper; use hyper::Server; use hyper::server::Request; use hyper::server::Response; fn hello(req: Request, res: Response) { let response = format!("Hello from {}!", req.remote_addr.ip()); res.send(response.as_bytes()).unwrap(); } fn main() { Server::http("127.0.0.1:3000").unwrap() .handle(hello).unwrap(); } Output from wrk -t8 -c400 -d60s http://localhost:3000/ http://localhost:3000/ Running 1m test @ http://localhost:3000/ 8 threads and 400 connections Thread Stats Avg Stdev Max +/- Stdev Latency 173.31us 88.10us 0.98ms 65.18% Req/Sec 35.06k 13.46k 45.48k 74.99% 8383194 requests in 1.00m, 775.50MB read Socket errors: connect 0, read 0, write 229, timeout 0 Requests/sec: 139487.00 Transfer/sec: 12.90MB According to top, rust_http is using about 2.8mb ram, 630% cpu, and wrk is using 700k and 370% cpu. Though I'm on completely different hardware, but the thing definitely screams... For reference: AMD Ryzen R9 3950X, 64GB DDR4@3200(CL18) under WSL2-Ubuntu (Windows 10 insiders)
sk0g 6y ago
Wouldn't processing time depend on your CPU, and unless you run the benchmarks in the article, this comparison would be meaningless?
- tracker1 6y ago
  Don't have a go environment setup, but could do that. The executable size and memory use was also lower.
  - sk0g 6y ago
    Well to be honest, Go binary sizes are always going to be big, to the benefit of ease of development and deployment. Memory usage, I don't know. The Fibonacci benchmark code was a bit... shit. Do you want to send over your Rust code in a Gist or something, and I can compile and compare?
    - tracker1 6y ago
      It's in the comment above...