Open kirawi opened 3 years ago
Some folks have been working on benchmarking infrastructure using Sightglass with possibly some CI integration and/or historical views -- @jlb6740 what's the latest status on that?
It doesn't look like it benchmarks the native equivalent.
In the past, at least in presentations I've seen, Sightglass results have included performance ratios against the same code running natively; and I think that's the plan going forward, but @jlb6740 or @fitzgen could comment further?
The benchmark suite is explicitly interally focused (is this PR a regression/improvement? what is our perf trend over time?) and we de-emphasized external presentation, "benchmarketing", and direct comparisons to other wasm engines or native, per the RFC.
@kirawi You could probably get the benchmarks running on native without a ton of work if you're motivated to do that, but it isn't really the focus of our benchmarking efforts.
I think it would still be valuable to have benchmarks against native code to help in deciding whether a project like Wasmtime fits design requirements. Wasm will obviously be overall slower, but how slow is a big deciding factor.
Feature
I would like it if Wasmtime would publish benchmarks of native Rust code versus it compiled down to Wasm.
Benefit
It would help decide how performant it is compared to a direct scripting language, and if its portability outweighs its worse performance. There isn't a source truth for such comparisons AFAIK except https://00f.net/2021/02/22/webassembly-runtimes-benchmarks/
Implementation
Outside of my knowledge.
Alternatives
Don't implement/publish such benchmarks.