Compare throughput benchmarks from various Bun HTTP framework

Test method: Average throughput

1. Get (/)
   • Request to [GET] /
   • Return hi
   • Headers must contains text Content-Type: text/plain, additional context is acceptable eg. Content-Type: text/plain; charset=utf-8
2. Params, query & header
   • Request to [GET] /id/:id
   • Extract path parameter, query string and setting headers.
   • For this benchmark, the request URL will be send as: /id/1?name=bun
   • Headers must contains x-powered-by to benchmark
   • Expected response: "1 bun" (${id} ${query})
     • You MUST NOT use hardcode string or index to extract querystring.
     • In a real-world situation, there's no enforcement that the request will follow the specification, using hardcode index to extract name=bun querystring will be prone to error.
     • To test if it pass the requirement, the implementation should be able to extract querystring dynamically (please treat the value of 'name=bun' can be any value beside 'bun', for example 'alice', 'hina'), which means that the same code should be able to extract querystring, for example:
     • /id/1?name=bun&id=1 -> should return 1 bun not 1 bun&id=1
     • /id/1?id=1 -> should return 1
     • Query beside name maybe not need to be extracted and is optional
   • Headers must contains text Content-Type: text/plain, additional context is acceptable eg. Content-Type: text/plain; charset=utf-8
3. Post JSON
   • [POST] /json
   • Mirror body to response
   • For the benchmark, the request body will be sent as: { "hello": "world" }
   • Expected response: { "hello": "world" }
   • Headers must contains text Content-Type: application/json, additional context is acceptable eg. Content-Type: application/json; charset=utf-8.

• bombardier
• Nodejs
• Deno
• Bun

bun run benchmark

Dump result will be available at results/[benchmark-name].txt

This benchmark is tested under the following condition:

• Windows 11 under WSL Debian
• Intel I7-13700K, DDR5 32GB 5600MHz
• Windows 11 22H2 build 22621.1778
• Debian GNU/Linux 11 (Bullseye), kernel: 5.15.90.1-microsoft-standard-WSL2
• Bun 1.0.0
• Node 18.16.0
• Deno 1.36.0

Tested on 8 Sep 2023 23:39 (GMT+7)

For results suffix with -node means that the framework is run in Node, otherwise is using Bun.

These results are measured in req/s:

Crash: | express (bun) | 512,117.39 | 513,848.37 | 511,647.49 | 510,856.31 | | cheetah (deno) | 300,995.98 | 507,117.96 | 378,810.25 | 17,059.73 |

See more detail in results

I highly recommended testing this benchmark on your machine yourself as performance in likely to vary between machine.