如何估算网站RPS峰值？

2018-06-05 by dongnan

目标

估算网站RPS峰值

概念

PV: page view 即页面被浏览的次数。
RPS: Requests per second 即每秒请求次数。

二八定律

峰值：按照二八定律的思想，网站每天 80%的访问量集中在 20%的时间里，这段20%时间叫做峰值时间。
公式：峰值RPS = (总PV数 * 80%) / (每天秒数 * 20%) 。
资源：简单来理解就是服务器，服务器 = 峰值RPS / 单台机器的最大RPS 。

举个栗子

假设，某项目网站每天有 1000万PV访问量，设计这个网站需要多少峰值RPS？

(10000000 * 0.8) / (86400 * 0.2) = 463/s

再假设，经过压力测试单台服务器最大RPS为 100/s，需要多少台服务器？

答案是大约5台 463 / 100 = 5

再再假设，考虑整个网站系统设计具有30%的冗余性，那么需要多少台服务器？

答案是大约7台 5 + 5 * 0.3

压力测试

使用 apache 压力测试工具，对某项目网站系统进行压测，共4次每次间隔 5分钟。

测试过程

第1次测试，请求:1000 并发:100 RPS:1048/s 平均等待时间:95.463/ms。

ab -n 1000 -c 100 https://www.demo.com/passport/login

# ...省略
Benchmarking www.demo.com (be patient)
Completed 100 requests
# ...省略
Completed 1000 requests
Finished 1000 requests

Server Software:
Server Hostname:        www.demo.com
Server Port:            443
SSL/TLS Protocol:       TLSv1.2,ECDHE-RSA-AES128-GCM-SHA256,2048,128

Document Path:          /passport/login
Document Length:        10668 bytes

Concurrency Level:      100
Time taken for tests:   0.955 seconds
Complete requests:      1000
Failed requests:        0
Write errors:           0
Total transferred:      11242000 bytes
HTML transferred:       10668000 bytes
Requests per second:    1047.53 [#/sec] (mean)
Time per request:       95.463 [ms] (mean)
Time per request:       0.955 [ms] (mean, across all concurrent requests)
Transfer rate:          11500.28 [Kbytes/sec] received

Connection Times (ms)
min  mean[+/-sd] median   max
Connect:        8   48  45.2     36     320
Processing:     9   43  11.7     45      94
Waiting:        9   38  11.1     40      90
Total:         19   91  45.9     83     361

Percentage of the requests served within a certain time (ms)
50%     83
# ...省略
99%    319
100%    361 (longest request)

第2次测试，请求:1000 并发:200 RPS:951.34/s 平均等待时间:210.231/ms 。

ab -n 1000 -c 200 https://www.demo.com/passport/login
# ...省略
Benchmarking www.demo.com (be patient)
Completed 100 requests
...省略
Completed 1000 requests
Finished 1000 requests

Server Software:
Server Hostname:        www.demo.com
Server Port:            443
SSL/TLS Protocol:       TLSv1.2,ECDHE-RSA-AES128-GCM-SHA256,2048,128

Document Path:          /passport/login
Document Length:        10668 bytes

Concurrency Level:      200
Time taken for tests:   1.051 seconds
Complete requests:      1000
Failed requests:        0
Write errors:           0
Total transferred:      11242000 bytes
HTML transferred:       10668000 bytes
Requests per second:    951.34 [#/sec] (mean)
Time per request:       210.231 [ms] (mean)
Time per request:       1.051 [ms] (mean, across all concurrent requests)
Transfer rate:          10444.26 [Kbytes/sec] received

Connection Times (ms)
min  mean[+/-sd] median   max
Connect:        7   88  80.8     61     519
Processing:     9   76  36.8     84     299
Waiting:        8   54  30.5     49     247
Total:         16  164  90.7    162     567

Percentage of the requests served within a certain time (ms)
50%    162
# ...省略
99%    399
100%    567 (longest request)

第3次测试，请求:1000 并发:300 RPS:806.40/s 平均等待时间:372.023/ms 。

ab -n 1000 -c 300 https://www.demo.com/passport/login
# ...省略
Benchmarking www.demo.com (be patient)
Completed 100 requests
# ...省略
Finished 1000 requests

Server Software:
Server Hostname:        www.demo.com
Server Port:            443
SSL/TLS Protocol:       TLSv1.2,ECDHE-RSA-AES128-GCM-SHA256,2048,128

Document Path:          /passport/login
Document Length:        10668 bytes

Concurrency Level:      300
Time taken for tests:   1.240 seconds
Complete requests:      1000
Failed requests:        0
Write errors:           0
Total transferred:      11242000 bytes
HTML transferred:       10668000 bytes
Requests per second:    806.40 [#/sec] (mean)
Time per request:       372.023 [ms] (mean)
Time per request:       1.240 [ms] (mean, across all concurrent requests)
Transfer rate:          8853.09 [Kbytes/sec] received

Connection Times (ms)
min  mean[+/-sd] median   max
Connect:        8  131 124.2     94    1009
Processing:     9  117  52.1    116     433
Waiting:        9   64  43.2     51     263
Total:         18  249 140.9    201    1041

Percentage of the requests served within a certain time (ms)
50%    201
# ...省略
99%    757
100%   1041 (longest request)

第4次测试，请求:5000 并发:100 RPS:589.81/s 平均等待时间:169.546/ms 其它: 出现 SSL握手失败。

ab -n 5000 -c 100 https://www.demo.com/passport/login
# ...省略
Benchmarking www.demo.com (be patient)
Completed 500 requests
# ...省略
Completed 3000 requests
SSL handshake failed (5).
# ...省略
Finished 5000 requests

Server Software:
Server Hostname:        www.demo.com
Server Port:            443
SSL/TLS Protocol:       TLSv1.2,ECDHE-RSA-AES128-GCM-SHA256,2048,128

Document Path:          /passport/login
Document Length:        10668 bytes

Concurrency Level:      100
Time taken for tests:   8.477 seconds
Complete requests:      5000
Failed requests:        1
(Connect: 0, Receive: 0, Length: 1, Exceptions: 0)
Write errors:           0
Total transferred:      56198758 bytes
HTML transferred:       53329332 bytes
Requests per second:    589.81 [#/sec] (mean)
Time per request:       169.546 [ms] (mean)
Time per request:       1.695 [ms] (mean, across all concurrent requests)
Transfer rate:          6473.94 [Kbytes/sec] received

Connection Times (ms)
min  mean[+/-sd] median   max
Connect:        0  130 305.0     13    3019
Processing:     5   29  47.7     17     823
Waiting:        8   27  41.7     16     622
Total:          5  160 310.5     34    3035

Percentage of the requests served within a certain time (ms)
50%     34
# ...省略
99%   1246
100%   3035 (longest request)

测试小结

压测端与服务器端，在相同的网络，所以网络带宽不是瓶颈。
RPS 随着并发数的增加而减少，系统负载升高、CPU成为瓶颈。
平均等待时间，并发数成正比，即并发数量越大，用户等待时间越长，直到网站系统拒绝响应(SSL握手失败)。

RPS测试原则

不同系统对最大负载的要求不一样。就通常的网站来说基本指标是：在合理的响应时间内，系统能提供最大的每秒请求数（QPS/RPS/TPS）。

合理响应时间，3-5s是一个合格的网站渲染首屏最多花费的时间，如果单纯看最大qps而不管响应时间的话，这个测试是无意义的。
每个页面包含的功能不一样多，所以指标也定不相同，那么访问最频繁的页面的性能指标应该是我们最应该关注的。
最重要的一点是，刚创业的站点应多注意快速开发业务，心里有性能这根弦，不犯愚蠢错误就足够了。等站点真正做出来再回头调优，应该是个甜蜜的过程。