How Fast is Ruby 2.5.0?
/Back in November, I posted speed results for Ruby 2.5.0 preview 1. It was barely faster than Ruby 2.4, which was a bit of a disappointment. However, one very important performance patch landed before it finished, which made a big difference in the final speed.
How big? Let's see, shall we?
Quick Graphs
You just want to see the graphs, I'll bet. I'm the same way. Here's a great start: total-time runs for Rails Ruby Bench. This measures the time taken to push a mixture of Discourse (Rails) requests through a big concurrent server:
Yup, those bars on the right are shorter.
So: not bad. What's that look like as a table of numbers and percent faster?
| Percentile | Ruby 2.4.3 | Ruby 2.5.0 | % Faster |
|---|---|---|---|
| 0% | 29.17 | 26.99 | 7.5% |
| 10% | 32.25 | 30.73 | 4.7% |
| 50% | 33.98 | 32.39 | 4.7% |
| 90% | 35.15 | 33.37 | 5.1% |
| 100% | 36.77 | 35.62 | 3.1% |
What's interesting here is that the higher (slower) runs gained less speed, which isn't usually how it works. That's almost certainly because of the unusual nature of the performance patch that was nearly all the speed difference: it was a more-or-less constant overhead per Ruby bytecode operation. If the slower runs had instructions that each took longer (pretty likely) then you'd expect them to gain less performance. Which is roughly what you see here.
If we asked, "sure, but what's the overall number? How much faster is Ruby 2.5.0?," I'd probably wind up answering in throughput, not the percentile per-run time. So let's look at throughput:
| Measurement | Ruby 2.4.3 | Ruby 2.5.0 | % Faster |
|---|---|---|---|
| Mean Throughput | 170.6 | 179.3 | 5.1% |
| Median Throughput | 171.0 | 179.6 | 5.0% |
How much faster? 5% faster for throughput on a big concurrent Rails server. Tell your friends!
Can it be faster than that? Sure. With a lot of small, faster operations I was seeing up to 7.5% faster on Rails requests, and Koichi was seeing up to 12% faster on some benchmarks.
But for the simple answer, "will it make my Rails app faster?" Yes. About 5% faster. Which isn't bad for a pretty calm upgrade that isn't likely to break anything. It'll just speed up your code and add a few nice features!