There is no benchmark Compared to tokio and async_std?

[zhuxiujia]

There is no benchmark Compared to tokio and async_std?

[notgull]

Benchmarking async runtimes is really hard. The potential workloads are so diverse that, at a point, it doesn't really matter.

As a general rule, there are two main bottlenecks in an async runtime: the reactor and the executor. async-std and smol use the same executor. tokio's executor is probably faster for longer term workloads (like if a network connection slowly dispatches packets of bytes over a long period of time), while smol's is better for shorter term workloads (like all the bytes being immediately available from a socket).

smol's reference executor is almost certainly slower than tokio's executor. But it's easy to slot another executor in (like smolscale).

These are just qualitative estimates based on the architectures of the runtimes. If the tokio team would like to collaborate on a more quantitative set of benchmarks I would be happy to oblige.

[varbhat]

I benchmarked axum powered by tokio vs axum powered by smol. wrk tool was used to benchmark both.

The top section of the image is benchmark of axum powered by tokio. The bottom section of the image is benchmark of axum powered by smol.

The axum endpoint just returns static string "Hello world". It's just homepage example of axum.

CC: @notgull

TLDR:
axum powered by tokio =>
Reqs/sec : 79.1k
Transfer/sec: 98.1MB

axum powered by smol =>
Reqs/sec: 67.5k
Transfer/sec: 83.6MB

[zhuxiujia]

I benchmarked axum powered by tokio vs axum powered by smol. wrk tool was used to benchmark both.

The top section of the image is benchmark of axum powered by tokio. The bottom section of the image is benchmark of axum powered by smol.

The axum endpoint just returns static string "Hello world". It's just homepage example of axum.

CC: @notgull

TLDR: axum powered by tokio => Reqs/sec : 79.1k Transfer/sec: 98.1MB

axum powered by smol => Reqs/sec: 67.5k Transfer/sec: 83.6MB

This doesn't seem as good as Tokio

[notgull]

The default executor for smol is quite slow. smolscale probably has better performance characteristics.

[sirius-cc-wu]

https://zenoh.io/blog/2022-04-14-rust-async-eval/