使用PHP驱动的MongoDB的单点查询性能测试

之前的测试，发现MongoDB的插入操作在使用PHP驱动时比使用mongo控制台更快。测试发现查询也是类似的情况（以下测试都使用WT引擎）。
http://blog.chinaunix.net/uid-20726500-id-4998173.html

1. 插入数据

-bash-4.1$ cat test.php
<?php
header("Content-Type:text/html;charset=utf-8");
//MongoDB有用户名密码并指定数据库admin
$conn = new Mongo("mongodb://127.0.0.1:27017/benchmark");

$db = $conn->admin;
//定制结果集(表名things)
$collection = $db->members;
//$time1 = xdebug_time_index();

for($i=1;$i<1000000;$i++){
$user = array('uname' => 'chuchuchu_'.$i, 'name' => '褚褚褚', 'password' => 'e10adc3949ba59abbe56e057f20f883e', 'email' => '[e-mail]');
$collection->insert($user);
}

$conn->close();
//$time2 = xdebug_time_index();
//echo "MongoDB响应时间为：".($time2-$time1)."秒";
?>

-bash-4.1$ time php test.php


real    1m45.062s
user    0m22.609s
sys    0m12.569s

2. 建索引

-bash-4.1$ mongo benchmark
...
> use admin
switched to db admin
> db.members.ensureIndex({uname:1})
{
    "createdCollectionAutomatically" : false,
    "numIndexesBefore" : 1,
    "numIndexesAfter" : 2,
    "ok" : 1
}

3.单点索引查询

-bash-4.1$ cat test_sel.php
<?php
header("Content-Type:text/html;charset=utf-8");
$conn = new Mongo("mongodb://127.0.0.1:27017/benchmark");

$db = $conn->admin;
$collection = $db->members;

for($i=1;$i<1000000;$i++){
$query = array( 'uname' => 'chuchuchu_'.$i);
$cursor = $collection->find( $query );
while( $cursor->hasNext() ) {
    $var=$cursor->getNext();
}
}

$conn->close();

?>

-bash-4.1$ time php test_sel.php

real    1m42.998s
user    0m22.888s
sys    0m13.160s

top的结果

[root@hanode1 ~]# top
top - 03:27:45 up 8 days, 8:04, 6 users, load average: 0.35, 0.35, 0.17
Tasks: 152 total, 1 running, 151 sleeping, 0 stopped, 0 zombie
Cpu(s): 18.7%us, 5.2%sy, 0.0%ni, 74.2%id, 0.0%wa, 0.0%hi, 2.0%si, 0.0%st
Mem: 1019320k total, 952192k used, 67128k free, 155224k buffers
Swap: 2064376k total, 184k used, 2064192k free, 332756k cached

PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
24809 postgres 20 0 575m 294m 11m S 65.2 29.6 3:23.02 mongod
28124 postgres 20 0 208m 7016 4528 S 34.9 0.7 0:02.05 php

索引查询速度大约每秒1w次，mongod消耗CPU 65.2%，经过计算将1个cpu core用满可以达到1.5w/s（1/0.652）。

4. 一级索引的测试

上面是二级索引的测试，那么"_id"的一级索引，性能会不会还有提升空间呢？也测一下看看。

4.1 先把之前是数据删掉。

-bash-4.1$ mongo benchmark
...
> use admin
switched to db admin
> db.members.drop()
true

4.2 插入数据

-bash-4.1$ cat test_2.php
<?php
header("Content-Type:text/html;charset=utf-8");
//MongoDB有用户名密码并指定数据库admin
$conn = new Mongo("mongodb://127.0.0.1:27017/benchmark");

$db = $conn->admin;
//定制结果集(表名things)
$collection = $db->members;
//$time1 = xdebug_time_index();

for($i=1;$i<1000000;$i++){
$user = array('_id' => 'chuchuchu_'.$i, 'name' => '褚褚褚', 'password' => 'e10adc3949ba59abbe56e057f20f883e', 'email' => '[e-mail]');
$collection->insert($user);
}

$conn->close();
//$time2 = xdebug_time_index();
//echo "MongoDB响应时间为：".($time2-$time1)."秒";
?>

-bash-4.1$ time php test_2.php


real    1m43.274s
user    0m21.286s
sys    0m12.985s

原来test.php中的'uname'被改成了'_id'

4.3 查询数据

点击(此处)折叠或打开

-bash-4.1$ cat test_sel_2.php
<?php
header("Content-Type:text/html;charset=utf-8");
$conn = new Mongo("mongodb://127.0.0.1:27017/benchmark");

$db = $conn->admin;
$collection = $db->members;

for($i=1;$i<1000000;$i++){
$query = array( '_id' => 'chuchuchu_'.$i);
$cursor = $collection->find( $query );
while( $cursor->hasNext() ) {
    $var=$cursor->getNext();
}
}

$conn->close();

?>

-bash-4.1$ time php test_sel_2.php

real    1m19.557s
user    0m21.236s
sys    0m12.156s

原来test_sel.php中的'uname'被改成了'_id'

查询top结果

[root@hanode1 ~]# top
top - 03:40:51 up 8 days, 8:17, 6 users, load average: 0.21, 0.37, 0.28
Tasks: 152 total, 2 running, 150 sleeping, 0 stopped, 0 zombie
Cpu(s): 17.5%us, 5.5%sy, 0.0%ni, 74.9%id, 0.0%wa, 0.0%hi, 2.1%si, 0.0%st
Mem: 1019320k total, 787364k used, 231956k free, 83380k buffers
Swap: 2064376k total, 184k used, 2064192k free, 142384k cached

PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
24809 postgres 20 0 711m 420m 11m R 58.2 42.2 6:02.83 mongod
29480 postgres 20 0 208m 7016 4528 S 41.6 0.7 0:02.78 php

索引查询速度大约每秒1.26w次，mongod消耗CPU 58.2% ，经过计算将1个cpu core用满可以达到2.2w/s（1/0.582）。

5. 总结

在我的测试环境下，假设mongod 用满1个CPU core，一级索引的单点查询QPS可达到2.2w/s，二级索引则大约是1.5w/s。