golang协程和python协程对比
golang协程和python协程对比，主要是测试在生产中的并发、效率等。
因为机器配置较弱（没钱开机器😄），均在load15负载约降为0的时候开启下次测试，避免机器负载影响。 同时程序无复杂逻辑处理，仅做返回。
go协程和python协程测试环境
压测环境
AWS 1核1G CPU为 E5-2676 v3 @ 2.40GHz，压测工具：wrk
目标环境
AWS 1核1G CPU为 E5-2676 v3 @ 2.40GHz，机器无弹性扩容性能，固死频率
前段均用nginx做反向代理
- golang环境：nginx->docker->golang->gin
- python环境：nginx->docker->python3.7->flask+gevent单进程模式
go协程测试结果
|
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 |
root@aws1:~# wrk -t6 -c200 -d30s http://x.x.x.x/ua Running 30s test @ http://x.x.x.x/ua 6 threads and 200 connections Thread Stats Avg Stdev Max +/- Stdev Latency 94.19ms 128.56ms 1.12s 96.09% Req/Sec 448.99 81.79 680.00 70.00% 80543 requests in 30.04s, 13.67MB read Requests/sec: 2681.19 Transfer/sec: 465.95KB root@aws1:~# wrk -t3 -c100 -d30s http://x.x.x.x/ua Running 30s test @ http://3.0.3.0/ua 3 threads and 100 connections Thread Stats Avg Stdev Max +/- Stdev Latency 34.64ms 5.17ms 65.64ms 94.30% Req/Sec 0.96k 69.42 1.05k 87.78% 85742 requests in 30.03s, 14.55MB read Requests/sec: 2854.81 Transfer/sec: 496.12KB |
golang在机器负载高的时候响应很慢，约为100ms左右，机器负载低的时候响应很快34ms。平均水平QPS约为2800(1核1G内存，2.4GHZ处理器)
python协程测试结果
|
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 |
root@aws1:~# wrk -t6 -c200 -d30s http://x.x.x.x/cppla Running 30s test @ http://x.x.x.x/cppla 6 threads and 200 connections Thread Stats Avg Stdev Max +/- Stdev Latency 186.31ms 38.63ms 1.21s 98.70% Req/Sec 182.58 93.09 333.00 56.87% 32052 requests in 30.05s, 6.85MB read Requests/sec: 1066.56 Transfer/sec: 233.28KB root@aws1:~# wrk -t3 -c100 -d30s http://x.x.x.x/cppla Running 30s test @ http://x.x.x.x/cppla 3 threads and 100 connections Thread Stats Avg Stdev Max +/- Stdev Latency 94.45ms 7.26ms 127.97ms 92.59% Req/Sec 350.32 42.29 450.00 77.56% 31425 requests in 30.04s, 6.71MB read Requests/sec: 1046.22 Transfer/sec: 228.81KB root@aws1:~# wrk -t3 -c50 -d30s http://x.x.x.x/cppla Running 30s test @ http://x.x.x.x/cppla 3 threads and 50 connections Thread Stats Avg Stdev Max +/- Stdev Latency 45.89ms 4.44ms 78.93ms 95.25% Req/Sec 349.89 41.41 454.00 80.67% 31381 requests in 30.03s, 6.70MB read Requests/sec: 1044.85 Transfer/sec: 228.51KB |
可以看出python在并发和机器负载低情况下跟golang响应方面差别不大，平均水平QPS约为1000(1核1G内存，2.4GHZ处理器)。不过并发能力确实弱很多，而且随着负载增高，越来越慢（还是动态语言解释的原因），不过话说instagram都用来处理上亿的并发，这点量算什么😄，Refer：《Flask gevent 多进程WSGI，每秒1w5 QPS》。