Rust sorting benchmark

[mo8it]

On my machine, Rust has almost the same performance now. Please update your blog post for a better comparison.

Why would you use the rand crate for random number generation and not rayon for parallelism?

rayon is written in 100% Rust. About the "Standard library" thing: Rust by design keeps its standard library as small as possible. Major crates like rayon are often maintained by the Rust maintainers themself. For rayon, the two main maintainers Josh Stone and Niko Matsakis are core Rust contributors. Niko is even a team leader of the language team: https://www.rust-lang.org/governance

[mppf]

@mo8it, thanks for making a Rust+Rayon version of the benchmark.

Why would you use the rand crate for random number generation and not rayon for parallelism?

Because the benchmark is about sorting, not about generating random numbers. So, the test programs can generate random numbers in any old way.

However, your comments here have pointed out to me that the article shouldn’t claim that the Chapel random number generation is 17x faster than one in Rust since the Rust code I measured was generating the random numbers with a crate (vs a standard library) and using different crates can make it faster (as you show here). So I'll update that piece.

Please update your blog post for a better comparison.

I agree that Rust+Rayon is an interesting comparison. At the same time, I think that we can all agree that Rayon isn’t in Rust’s standard library. There are all kinds of interesting sort implementations in the world that aren’t in standard libraries & these are not what the article is about.

In particular with Rust, as you say, there is a design goal to have a small standard library. But, in this case, I think that makes it harder for newcomers to use parallelism. For one thing, they have to start by choosing between Rayon and Crossbeam. I think that something being in a standard library makes it more valuable and accessible and that’s why the article focuses on standard libraries. Of course, you are free to disagree with my position on the importance of standard libraries.

That said, I’d like to include your Rust+Rayon benchmark among the other benchmarks as an alternative rather than a replacement for the original. If you want to help me do that, there will be some updates that are needed to match our contributing guidelines. Please see https://chapel-lang.org/docs/latest/developer/bestPractices/DCO.html#best-practices-dco for the most immediate step, and in addition, it’ll be necessary to change the PR to add a separate test/library/packages/Sort/performance/comparison/sort-random-rust-rayon/ directory to include your implementation.

[mppf]

I've created a Discourse topic for the blog post being discussed here: https://chapel.discourse.group/t/comparing-standard-library-sorts-the-impact-of-parallelism/30411 . That discourse group is a better place for discussion about the blog post itself.

[mo8it]

For one thing, they have to start by choosing between Rayon and Crossbeam.

Rayon and Crossbeam do completely different things and aren't interchangeable.

I signed my commits and separated the rayon sorting.