|
|
分为本地、线上测试，所有框架均用默认配置
本地测试环境: MacOS 12.7
测试样本 并发数300 持续10秒 每个框架执行3次，取平均值
测试命令 bombardier -c 300 -d 10s http://127.0.0.1:3000/
本地helloworld(删除多余的代码，只响应返回helloworld文本)
自己写的框架
==============>
Statistics Avg Stdev Max
Reqs/sec 80010.09 4575.56 89098.13
Latency 3.82ms 372.66us 43.54ms
HTTP codes:
1xx - 0, 2xx - 795573, 3xx - 0, 4xx - 0, 5xx - 0
others - 0
Throughput: 17.16MB/s
next.js框架
==============>
Statistics Avg Stdev Max
Reqs/sec 3755.82 515.57 5050.54
Latency 89.26ms 734.71ms 16.04s
HTTP codes:
1xx - 0, 2xx - 37618, 3xx - 0, 4xx - 0, 5xx - 0
others - 275
Errors:
dial tcp 127.0.0.1:3002: connect: connection reset by peer - 158
the server closed connection before returning the first response byte. Make sure the server returns 'Connection: close' response header before closing the connection - 74
timeout - 42
write tcp 127.0.0.1:57438->127.0.0.1:3002: write: broken pipe - 1
Throughput: 8.97MB/s
nuxt.js框架
==============>
Statistics Avg Stdev Max
Reqs/sec 2181.77 1577.13 5278.16
Latency 158.78ms 1.11s 16.24s
HTTP codes:
1xx - 0, 2xx - 21373, 3xx - 0, 4xx - 0, 5xx - 0
others - 322
Errors:
dial tcp 127.0.0.1:3001: connect: connection reset by peer - 195
the server closed connection before returning the first response byte. Make sure the server returns 'Connection: close' response header before closing the connection - 95
timeout - 29
write tcp 127.0.0.1:53730->127.0.0.1:3001: write: broken pipe - 1
write tcp 127.0.0.1:53728->127.0.0.1:3001: write: broken pipe - 1
write tcp 127.0.0.1:54196->127.0.0.1:3001: write: broken pipe - 1
Throughput: 1.46MB/s
--------------------------
--------------------------
本地真实网站，需要读取数据库，构建页面
自己写的框架
==============>
Statistics Avg Stdev Max
Reqs/sec 1082.09 1980.78 10934.21
Latency 271.90ms 27.26ms 465.08ms
HTTP codes:
1xx - 0, 2xx - 11111, 3xx - 0, 4xx - 0, 5xx - 0
others - 0
Throughput: 15.63MB/s
next.js
==============>
Statistics Avg Stdev Max
Reqs/sec 849.89 377.48 4222.65
Latency 889.52ms 3.79s 30.91s
HTTP codes:
1xx - 0, 2xx - 8524, 3xx - 0, 4xx - 0, 5xx - 0
others - 274
Errors:
dial tcp 127.0.0.1:45454: connect: operation timed out - 129
the server closed connection before returning the first response byte. Make sure the server returns 'Connection: close' response header before closing the connection - 109
dial tcp 127.0.0.1:45454: connect: connection reset by peer - 30
timeout - 5
write tcp 127.0.0.1:53003->127.0.0.1:45454: write: broken pipe - 1
Throughput: 7.05MB/s
以前用nuxt.js写的网站换了新的数据库，所以不测试
--------------------------
--------------------------
线上测试环境: CPU 1H(i9-11900K)1G的美国小鸡
线上测试helloworld
自己写的框架 cpu占用率最高4% 平均3.2%
==============>
Statistics Avg Stdev Max
Reqs/sec 1142.57 2227.97 15024.57
Latency 256.43ms 60.41ms 557.79ms
HTTP codes:
1xx - 0, 2xx - 11666, 3xx - 0, 4xx - 0, 5xx - 0
others - 0
Throughput: 260.31KB/s
next.js框架 cpu占用率最高48%，平均35%
==============>
Statistics Avg Stdev Max
Reqs/sec 1098.43 857.13 10456.58
Latency 269.57ms 124.37ms 2.47s
HTTP codes:
1xx - 0, 2xx - 11268, 3xx - 0, 4xx - 0, 5xx - 0
others - 0
Throughput: 4.15MB/s
nuxt.js框架 cpu占用率最高76%，平均68%
==============>
Statistics Avg Stdev Max
Reqs/sec 1097.11 852.92 8386.19
Latency 269.70ms 83.80ms 0.86s
HTTP codes:
1xx - 0, 2xx - 11250, 3xx - 0, 4xx - 0, 5xx - 0
others - 0
Throughput: 1.20MB/s
--------
----------------
----------------------
线上测试的请求性能接近(1.1w，这是cpu的瓶颈)，不过cpu占用率差异很大。
以前用nuxt.js、next.js来开发一个网站。
后来一些原因想仿写一个:
1. 网站运行期间莫名错误不知道哪来
2. 页面加载会有很多额外文件，臃肿
3. 页面上有nuxt或next的文字，不舒服
4. 类名太长，没有压缩
5. 构建时间太久
6. 对于简单网站大材小用了
于是写了一个简单的ssr框架，模仿了nuxt和next的文件目录结构，很好处理了上面的问题，相当于就是字符串拼接
性能高的原因是bun这个js运行时，内置web服务器是uwebsockets，性能非常的高。在techempower网站中js语言排名最高
不过bun发展没有很久，避免不了一些bug，遇到一个比较严重bug，挺不可思议，当使用Bun.listen返回大于130kb的数据，pc端能够正常接受，但是移动端就卡住，无响应。但是使用Bun.serve就没有这个问题。
当时反馈过，然后怀疑我有隐藏的换行符，我还没来得及解释就关闭了
关于原因第四点，我没测试之前认为类名压缩了能很好降低文件体积，诸如.main-left-content-a => .a 、.video-content-div => .b
不过还是想多了，顶多降低1~2kb大小，没我预期的多，看来gzip压缩比我想象中要牛逼
不过类名压缩能够隐藏原有类名的含义，还是有点保护性的 |
|
