Improve benchmarks

[rafaell-lycan]

Hi 👋

I've been using Hono for a couple of weeks now, and I gotta say it is simple and works!

One of the small things I haven't seen yet is a better benchmark comparison of a non-serverless environment vs. other Node web frameworks such as Express, Hapi, Koa, Restify, and Fastify on Fly.io or similar.

I came across this proposal after seeing these results and wondered if there's something odd with the tests or is due to Hono's nature of being optimized for serverless.

[yusukebe]

Hi @rafaell-lycan

In the Node environment, Hono is very slow because it uses an Adapter to convert between Web Standard APIs and Node input/output. Therefore, we do not really want to inform them of the results.

But, the results of this benchmark are weird. At least Hono running on Bun is faster than Fastify. Below are the results at hand.

Maybe there's a problem with the way it's set up.

[yusukebe]

Ah, in that benchmark, Fastify uses cluster to launch multiple processes. This is why it is faster than Hono.

[yusukebe]

The solution to this would be to make an adapter for Bun.

[rafaell-lycan]

Hey @yusukebe 👋 おはようございます 🌅

Thanks for dedicating time to this thread.

QQ: Does the adapter have any potential drawbacks that might be holding the performance or would be just due to the conversion?

Why: I knew that Hono would be faster than Express, but I was expecting somewhere near Restify & Koa performance.

This is definitely not the end of the world, and I bet things might get better in the near future by doing small optimizations on the Node adapter, while obviously when using Bun things will probably skyrocket.

I'm buying the whole initiative behind Hono and its simplicity (It reminds me Koa), and I guess that in the mid/long term having a more detailed benchmark (benchmark.md?) would be more explanatory on why using it and where it shines.

ありがとう

[rafaell-lycan]

The solution to this would be to make an adapter for Bun.

Is it worth tho? I'm running a simple server using a Dockerfile on Fly.io and it just works.

FROM oven/bun

WORKDIR /usr/src/app
COPY . .
RUN bun install
EXPOSE 3000

CMD ["bun", "start"]

[yusukebe]

Hey, good morning!

It is certainly better to have more benchmarks, but we want to make them as internally oriented as possible. This is because it is very difficult and trouble to publish and maintain benchmarks. Authors of other frameworks may complain about the results. Also, we need to keep them up-to-date. So, we would like to have external benchmarks as much as possible as a reference.

But this is an ideal and just my opinion.

[yusukebe]

Is it worth tho? I'm running a simple server using a Dockerfile on Fly.io and it just works.

No problem in that case. But, in that benchmark, Fastify uses multiple processes, because Hono was trying to do that with pm2, but was failing. We may need to customize it for Bun to make it work. Need to investigate.