Concurrency In Rust; Can It Stack Up Against Go’s Goroutines?

Concurrency In Rust; Can It Stack Up Against Go’s Goroutines?

The post Concurrency In Rust; Can It Stack Up Against Go’s Goroutines? appeared first on Qvault.

One of the primary goals of the Go programming language is to make concurrency simpler, faster, and more efficient. With Rust growing in popularity let’s see how its concurrency mechanisms stack up against Go’s.

A Refresher On Goroutines

In Go, concurrency is accomplished by spawning new goroutines:

package main

import (
    "fmt"
    "time"
)

func main() {
    go func() {
        for {
            fmt.Println("one second passed")
            time.Sleep(time.Second)
        }
    }()
    fmt.Println("waiting 10 secs for goroutine")
    time.Sleep(time.Second * 10)
}

In the example above, we use the go keyword to signify that we want to run the provided anonymous function in a goroutine. Execution at that point splits – execution continues on the main thread, but the runtime is now running the goroutine in parallel.

Goroutines are lightweight and take advantage of all of the processing power available. If two goroutines are running they will efficiently use at most two cores. If one hundred goroutines are running they will use at most one hundred cores, but can efficiently run on as few as one.

What About Rust?

In Rust there are two ways to run code concurrently. Async/Await, and threading. Async/Await allows a single thread (at most one CPU core) to asynchronously perform operations, while threading makes use of multiple cores and uses operating-system threads.

Async/Await

Let’s take a look at async first. and don’t forget to add the following dependencies to your project’s Cargo.toml:

[dependencies]
futures = "0.3.5"
async-std = "1.5.0"
use std::time::Duration;
use futures::executor::block_on;
use async_std::task;

fn main() {
    let future = async_main();
    block_on(future);
}

async fn async_main() {
    print_for_five("await 1").await;

    let async_one = print_for_five("async 1");
    let async_two = print_for_five("async 2");

    futures::join!(async_one, async_two);
}

async fn print_for_five(msg: &str) {
    for _ in 0..5 {
        task::sleep(Duration::from_secs(1)).await;
        println!("one second has passed: {}", msg)
    }
}

We start by creating a new async function, _async_main_, then use the _block_on_ function to block and execute _async_main_. Because _async_main_ is an asynchronous function, we are able to await other async functions inside of it, as well as execute them concurrently.

The first call in _async_main_ is an await on our async _print_for_five_ function which prints the message “ one second has passed: await 1 ” once each second for 5 seconds. Because we used the await keyword, _async_main_ will block and wait for _print_for_five_.

Next we create two futures (very similar to JavaScript promises) by calling _print_for_five_ anew with new messages. The futures do not begin execution until the next line where we use the join macro. Join executes the futures concurrently and blocks until all the futures have completed. Join is very similar to JavaScript’s PromiseAll.

The program will print the following:

one second has passed: await 1
one second has passed: await 1
one second has passed: await 1
one second has passed: await 1
one second has passed: await 1
one second has passed: async 1
one second has passed: async 2
one second has passed: async 1
one second has passed: async 2
one second has passed: async 1
one second has passed: async 2
one second has passed: async 1
one second has passed: async 2
one second has passed: async 1
one second has passed: async 2

Where the last ten lines are all printed within five seconds of each other because _print_for_five(“async 1”)_ and _print_for_five(“async 2”)_ were executed concurrently.

Rust’s Threading

As opposed to async/await, threading in Rust takes advantage of multi-core hardware. When a new thread is spawned, the operating system knows that these separate threads of the program can be executed in parallel on different cores at exactly the same time.

Let’s take a look at the following example:

use std::thread;
use std::time::Duration;

fn main() {
    thread::spawn(|| {
        for _ in 1..10 {
            println!("Hello after 1 second from the spawned thread");
            thread::sleep(Duration::from_millis(1000));
        }
    });

    for _ in 1..5 {
        println!("Hello after 1 second from the main thread");
        thread::sleep(Duration::from_millis(1000));
    }
}

We spawn a new thread using the standard library’s spawn function. The spawn function take a closure as its argument and executes it in parallel. As you can see by running the program, it only takes five seconds to print all ten statements because each thread is sleeping independently.

Which is Best?

You’ve heard it before but no approach is best, they are all just different. I argue that Go is the best at keeping it simple. Go provides only one method (goroutines) to achieve concurrency, and the syntax is elegant. Rust provides two methods which are tailored to different problems.

Goroutines vs Async/Await

Goroutines are very different from async/await. Async/Await can accomplish concurrency, but not parallelism. In other words, while async/await can logically execute two functions at once, it can’t practically do so because it uses at most one thread (one core).

Async/Await is a useful paradigm for programs that have heavy I/O wait times, but aren’t doing compute heavy workloads.

For example, with async/await one could efficiently make several database queries and respond to multiple HTTP requests at once. What async/await can’t do well (but threading and goroutines can handle without issue) is mine bitcoin while compressing images across a 4-core processor.

Goroutines vs Threading

Goroutines are more lightweight and efficient than operating-system threads. As a result, a program can spawn more total goroutines than threads. Goroutines also start and clean themselves up faster than threads due to less system overhead.

The big advantage of traditional threading (like that of Rust) over the goroutine model is that no runtime is required. Each Go executable is compiled with a small runtime which manages goroutines, while Rust avoids that extra fluff in the binary.

Thanks For Reading

Hit me up on twitter @wagslane if you have any questions or comments.

Follow me on Dev.to: wagslane

The post Concurrency In Rust; Can It Stack Up Against Go’s Goroutines? appeared first on Boot.dev.