AJP与HTTP比较和分析

系统环境：
OS:Ubuntu 10.10 (2G)
Servlet Container:tomcat-tomcat-7.0.23  (最大内存：default 256M  maxThreads:500)
Web server: apache2.2 (maxClient:250)
设置apache最大连接数

在/usr/local/etc/apache22/httpd.conf中加载MPM配置（去掉前面的注释）：
# Server-pool management (MPM specific)
Include etc/apache22/extra/httpd-mpm.conf

修改d-mpm.conf中对应module如下
<IfModule mpm_prefork_module>
                StartServers                      5
                MinSpareServers                   5
                MaxSpareServers                  10
                ServerLimit                     3000
                MaxClients                      2000
                MaxRequestsPerChild               0
</IfModule>]

一、benchmark测试方法
./ab -n 100000 -c 100 http://localhost/test
使用ab进行测试，当并发较高时，会出现以下错误
错误1：apr_socket_recv: Connection reset by peer (104)

apache2.2及以下版本
修改support下面的ab.c源代码， 大概在
line 1369， 修改成
1368                 return;
1369             } else {
1370                 //apr_err("apr_socket_recv", status);
1371                 bad++;
1372                 close_connection(c);
1373                 return;
1374             }
然后编译安装

错误2：Too many open files (24)

(PS；使用AJP协议和HTTP协议，多次进行测试，取测试结果较好的一次进行比较)

二、使用AJP后的benchmark

Benchmarking localhost (be patient)
Completed 10000 requests
Completed 20000 requests
Completed 30000 requests
Completed 40000 requests
Completed 50000 requests
Completed 60000 requests
Completed 70000 requests
Completed 80000 requests
Completed 90000 requests
Completed 100000 requests
Finished 100000 requests


Server Software:        
Server Hostname:        localhost
Server Port:            80

Document Path:          /test
Document Length:        347 bytes

Concurrency Level:      100
Time taken for tests:   50.270 seconds
Complete requests:      100000
Failed requests:        0
Write errors:           0
Total transferred:      48712175 bytes
HTML transferred:       34708675 bytes
Requests per second:    1989.24 [#/sec] (mean)
Time per request:       50.270 [ms] (mean)
Time per request:       0.503 [ms] (mean, across all concurrent requests)
Transfer rate:          946.29 [Kbytes/sec] received

Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:        0   23  15.3     21     714
Processing:     2   27  19.8     24     724
Waiting:        1   22  15.4     20     714
Total:          9   50  25.3     44     747

Percentage of the requests served within a certain time (ms)
  50%     44
  66%     49
  75%     53
  80%     55
  90%     64
  95%     73
  98%     87
  99%    101
 100%    747 (longest request)

三、使用HTTP 后的benchmark

Benchmarking localhost (be patient)
Completed 10000 requests
Completed 20000 requests
Completed 30000 requests
Completed 40000 requests
Completed 50000 requests
Completed 60000 requests
Completed 70000 requests
Completed 80000 requests
Completed 90000 requests
Completed 100000 requests
Finished 100000 requests


Server Software:        Apache-Coyote/1.1
Server Hostname:        localhost
Server Port:            80

Document Path:          /test
Document Length:        347 bytes

Concurrency Level:      100
Time taken for tests:   55.392 seconds
Complete requests:      100000
Failed requests:        0
Write errors:           0
Total transferred:      51415934 bytes
HTML transferred:       34710757 bytes
Requests per second:    1805.32 [#/sec] (mean)
Time per request:       55.392 [ms] (mean)
Time per request:       0.554 [ms] (mean, across all concurrent requests)
Transfer rate:          906.47 [Kbytes/sec] received

Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:        0   24  10.1     24     185
Processing:     8   31  10.7     28     215
Waiting:        1   26  10.0     24     198
Total:         13   55  12.3     51     247

Percentage of the requests served within a certain time (ms)
  50%     51
  66%     54
  75%     58
  80%     60
  90%     66
  95%     74
  98%     86
  99%     99
 100%    247 (longest request)

四、结论
If integration with the native webserver is needed for any reason, an AJP connector will provide faster performance than proxied HTTP.
前端apache，后端tomcat，通过ajp协议访问性能优于http协议，随着并发量的提升，效果会更加趋于明显。可以从吞吐率和总时间开销上观察。
（吞吐率：单位时间内计算机的处理请求来描述其并发处理能力）

可以参考下ajp协议的设计 http://httpd.apache.org/docs/2.2/mod/mod_proxy_ajp.html

原因总结如下：
1、ajp使用长连接保持webServer和servlet Container的通信，减少了建立tcp连接的开销。可以通过观察tomcat/manager 下serverStatus，ajp建立的连接都处于keepalive的状态。
2、ajp使用一定的协议格式，减少了传递的报文数据大小，节省了带宽。可以通过观察ajp和http 的benchmark报告重看到，Total transferred 一项有明显的区别。