Bun is 2x faster than Deno

[aquapi]

I have benchmarked Deno a few times on my machine and many benchmarks Deno is 2x slower than Bun
Maybe you can try use another benchmarking tool like wrk or bombardier to see the result.

Funding

• You can sponsor this specific effort via a Polar.sh pledge below
• We receive the pledge once the issue is completed & verified

[aquapi]

You can see these benchmarks

• bunsvr/benchmark
• SaltyAom/bun-http-framework-benchmark

[JasperVanEsveld]

There are quite a lot of benchmarks around and not a lot seem to agree with eachother, let alone that bun is 2x faster than deno.
For your bunsvr/benchmark, where can I see that bun is 2x faster?
I can only see different frameworks not runtimes in the results.

Looking at SaltyAom/bun-http-framework-benchmark there is an open issue that shows windows could impact the results.
The Linux results in that thread also wildly vary, one where node + uws wins (uws is also what bun uses).
Also the results from this user shows that deno with oak is within ~11%* of bun (edit: missed that JSON results were bugged, thanks @aquapi).
Even though in the results on this repo bun is 2.7x faster than deno with oak.
Like I said, the benchmarks don't seem to agree on much...

I am not involved with this repository whatsoever, but I think it would be appreciated to find a problem with their methodology instead of dropping a claim like this.
For example:

• Is there an issue with oha the load generator used, how would wrk or bombardier be better?
• Is there an issue with the code for either deno or bun?
• What OS is used?
• Is WSL (Windows Subsystem for Linux) used, if so does it have any impact on the results?
• What are the version numbers of the frameworks?

Which I think would all be valid questions and would be easier to follow up on.
Personally I would love to know version numbers and which OS is used for example.
This would also make it easier to validate these results.

[aquapi]

For simple benchmarking, wrk or wrk2 should be fine.
It provides a Lua API for you to work with and wrk is the most performant benchmarking tools.

The result mentioned above has an issue in POST /json test, which made the result inaccurate.
@JasperVanEsveld

[icetbr]

bun:  89,218.22 req/s
deno: 70,772.89 req/s
node: 28,542.56 req/s

Intel(R) Core(TM) i7-8565U CPU @ 1.80GHz
Linux Mint 21.2 kernel 5.15.0-82-generic

node v20.3.1
bun 1.0.0
deno 1.36.1 (release, x86_64-unknown-linux-gnu) v8 11.6.189.12 typescript 5.1.6

deno.ts

Deno.serve({ port: 3000 }, () => new Response("Hello, Bench!"));

bun.ts

export default { port: 3000, fetch: () => new Response('Hello, Bench!') };

node.js

require('http').createServer((request, response) => (response.write('Hello, Bench!'), response.end())).listen(3000);

bombardier --fasthttp -c 500 -d 10s http://localhost:3000/

All the best deno frameworks stay around 67,689.84 req/s (oak)

[aquapi]

WSL may have some impact on the result as it has some limitation as a virtual machine.
Idk if Deno has improved or not, but still.

I suggest you wrap the node handler with setImmediate and see the result

[icetbr]

Like this? No difference

const handler = response => () => {
    response.write('Hello, Bench!');
    response.end();
}
require('http').createServer((request, response) => setImmediate(handler(response))).listen(3000);

[aquapi]

@icetbr test on MacOS or other distro than Linux Mint
Bun seems to be performing bad in Linux Mint for some reasons?
(I suggest using Ubuntu)

[icetbr]

Mint is Ubuntu. I don't have MacOS, sorry. My benchmarks seems consistent with some articles I found online by others. Post your results if you can.

[aquapi]

@icetbr bunsvr/benchmark is mine

[icetbr]

Cool! I can only see bun frameworks for now, are you updating the repo?

[aquapi]

@icetbr Can you give me a framework code because I don't want to install Deno lint

[icetbr]

Which framework? Deno itself? I've used the ones from here
https://github.com/SaltyAom/bun-http-framework-benchmark/tree/main/src/deno

And pure deno

Deno.serve({ port: 3000 }, () => new Response("Hello, Bench!"));

[aquapi]

@eliassjogreen Why are benchmark results flunctuating on every single update?
You should check the benchmarking tool again or use wrk2.

[aquapi]

@icetbr Bun performs better in Ubuntu than in Linux Mint

[icetbr]

Do you have the stats? That would be unusual, as I'm not aware of any performance difference between any program between any linux distro.

[aquapi]

@icetbr if you have Mint you can for example test node:test with bun:test.
On my ubuntu machine bun is 3x faster
It may varies between distros as I see other results on Mint states that node:test is 3x faster

[icetbr]

Do you have any reference of this? It makes no sense, its like saying strawberry ice cream is colder then chocolate ice cream. Mint IS Ubuntu. Even Arch, which is way different, there is no performance difference. I don't doubt your results, but I'm curious of the circumstances of the test.

Even MacOs has the same performance. At least it used to be, I stopped following the benchmarks of dev environments a while ago, when i was deciding between a PC vs Mac for Node development.

[aquapi]

@icetbr See strager YT video

[Shyam-Chen]

https://github.com/Vanilla-IceCream/fastify-uws?tab=readme-ov-file#benchmarks

	Version	Language	Router	Requests/sec
bun	1.0.33	TypeScript/Bun	✗	157,740.9324
deno	1.41.3	TypeScript/Deno	✗	124,009.3235
node	20.11.1	JavaScript/Node	✗	78,002.1550

Bun is 2x faster than Node, not Deno.

[aquapi]

@Shyam-Chen It was