perf(tokio): add improved runtime thread handling - multi cpu

[j-mendez]

• improve multi cpu architecture performance
• add improved thread runtime performance

-- note

This PR leverages multi-core CPUs and improves the spawn capabilities for operations. By default threads live for 10 seconds after idle which makes sense for re-using the same thread except for applications that can be ran once and exited - non http servers etc.

[CLAassistant]

All committers have signed the CLA.

[ry]

Thanks for the patch! Do you have some benchmark that demonstrates the improvement?

[j-mendez]

@ry you are welcome! I can get a benchmark this week. Need to make a minimal demo since the project is private, should be straight forward for the example. Thank you for the JS runtime using Rust!

[j-mendez]

@ry you are welcome! I can get a benchmark this week. Need to make a minimal demo since the project is private, should be straight forward for the example. Thank you for the JS runtime using Rust!

Busy week - made a little progress on the benchmarks will try to get it done soon. Feel free to take this into other hands if moving too slow!

[j-mendez]

@ry pardon the delay, I went ahead and attached benchmarks for the runtime changes. Thank you for the patience.

[littledivy]

@j-mendez Thank you for the benchmarks. Although none of them are doing anything asynchrnous so it's unlikely that we are measuring spawn / spawn_blocking performance here. A better benchmark would be Deno.writeFile or Deno.readFile which uses spawn_blocking for file read/writes.

[j-mendez]

@j-mendez Thank you for the benchmarks. Although none of them are doing anything asynchrnous so it's unlikely that we are measuring spawn / spawn_blocking performance here. A better benchmark would be Deno.writeFile or Deno.readFile which uses spawn_blocking for file read/writes.

You are welcome, I appreciate the lead on the next steps. I am going to see if I can look into that since running the bench at root is a bit harder to track due to the amount of benches ran.

[littledivy]

@j-mendez Try this one deno run -A --unstable cli/bench/write_file.js. It tries to write small chunks to /dev/null as fast as possible.

[j-mendez]

@j-mendez Try this one deno run -A --unstable cli/bench/write_file.js. It tries to write small chunks to /dev/null as fast as possible.

I attached the logs of the before and after. I noticed afterwards my rust-analyzer ran in between reason for the drastic differences on branch checkouts.

[littledivy]

Ran the microbenchmarks again and i see no difference. I'd prefer to stay to the default unless there is a clear perf improvement. Thank you for the PR!