Elysia is defeated by old js frameworks. about elysia HOT 15 CLOSED

new result: https://web-frameworks-benchmark.netlify.app/result?l=javascript

And my personal analysis:

1. uwebsockets.js is excellent. It defeats most web frameworks and let js (EcmaScript) be comparable to native languages such as go, rust and java.
2. The frameworks based on uwebsockets.js (elysia, hono, stricjs, hyper-express, etc) introduce performance loss. Hope it could be minimized in the future.
3. HappyX is the champion of the benchmark. Just wondering what does it do.

from elysia.

Could you open PR for pm2 @jim-king-2000 ?

We use keep alive @SaltyAom
https://github.com/the-benchmarker/web-frameworks/blob/bd09015eb57e9b0b2aab5dbee2ed577671275caf/.tasks/collect.rake#L70

from elysia.

This is kind of suspicious, normally framework that runs on Bun should technically be faster than Node since Bun is based on WebSocket.
Digging on to that.

from elysia.

I think because of this:

Hmm, I dig it a bit, my guess is that since Node is using pm2-runtime, it's spawning multi-process to run multiple servers in parallel.
Fast, Deno-based one is also slower as well.
HyperExpress couldn't outperform Fiber multithreaded without this as well.
So kinda make sense.

https://twitter.com/saltyAom/status/1681697389189668864?s=20

You can see on this PR that Node is using pm2-runtime.

https://github.com/the-benchmarker/web-frameworks/pull/6422/files

So basically because Node framework uses multi-process, (using multiple threads to spawn servers simultaneously) therefore, Elysia is slower because, on that benchmark, any Bun-based framework uses only single-thread, therefore not quite fair.

from elysia.

Thank you very much for the analysis and quick response. How to modify the benchmark to create a fair comparison?

from elysia.

Hi @jim-king-2000 @SaltyAom ,

The purpose of the benchmark you notice is to reflect common usage in real workloads.

It's a common practice to use pm2 on top of node in order to spawn multiple processes.

Is there not such a thing is bun and deno communities?

from elysia.

So basically because Node framework uses multi-process, (using multiple threads to spawn servers simultaneously) therefore, Elysia is slower because, on that benchmark, any Bun-based framework uses only single-thread, therefore not quite fair.

Do you mean that we should also run bun relevant tests with pm2-runtime?

from elysia.

Maybe we should use pm2 start app.ts --interpreter="deno" --interpreter-args="run --allow-net for deno @jim-king-2000

Not sure about practices for running bun on production @SaltyAom

from elysia.

Please apply the change to both deno and bun frameworks. And I'm looking forward to the new benchmark result.

from elysia.

Not sure about practices for running bun on production @SaltyAom

Also use pm2.

from elysia.

Could you provide me both command ?

from elysia.

It's fine to use PM2 to run Bun server.

For Bun, I would recommend using this command.

pm2-runtime start "NODE_ENV=production bun src/index.ts" -i max

As for Deno, I think @waghanza command would also be working fine.

from elysia.

Also, this is from Jarred (Creator of Bun), may I ask this question as well?

He suggests that the slower response might be due to the usage of HTTP 1.0 or Keep Alive being disabled.

from elysia.

Could you open PR for pm2 @jim-king-2000 ?

Sure.

from elysia.

I turned the benchmark of bun and deno to multi-processing. Let's see the new result.

from elysia.