Apache在配置编译时可以自主的选择想要使用的MPM模块,使用./configure --with-mpm=MPM命令。我们主要了解prefork和worker这两种MPM模块。
Prefork
如果不用“--with-mpm”显式指定某种MPM,prefork就是Unix平台上缺省的MPM。它所采用的预派生子进程方式,用单独的子进程来处理不同的请求,进程之间彼此独立。在make编译和make install安装后,使用httpd -l来确定当前使用的
MPM是prefork.c。查看httpd-mpm.conf配置文件,里面包含如下默认的配置段:
代码如下
<IfModule prefork.c>
StartServers 5
MinSpareServers 5
MaxSpareServers 10
MaxClients 150
MaxRequestsPerChild 0
</IfModule>
prefork控制进程在最初建立“StartServers”个子进程后,为了满足MinSpareServers设置的需要创建一个进程,等待一秒钟,继续创建两个,再等待一秒钟,继续创建四个……如此按指数级增加创建的进程数,最多达到每秒32个,直到满足MinSpareServers设置的值为止。这种模式可以不必在请求到来时再产生新的进程,从而减小了系统开销以增加性能。MaxSpareServers设置了最大的空闲进程数,如果空闲进程数大于这个值,Apache会自动kill掉一些多余进程。这个值不要设得过大,但如果设的值比MinSpareServers小,Apache会自动把其调整为MinSpareServers+1。如果站点负载较大,可考虑同时加大MinSpareServers和MaxSpareServers。MaxRequestsPerChild设置的是每个子进程可处理的请求数。每个子进程在处理了“MaxRequestsPerChild”个请求后将自动销毁。0意味着无限,即子进程永不销毁。虽然缺省设为0可以使每个子进程处理更多的请求,但如果设成非零值也有两点重要的好处:1、可防止意外的内存泄漏。2、在服务器负载下降的时侯会自动减少子进程数。因此,可根据服务器的负载来调整这个值。MaxClients是这些指令中最为重要的一个,设定的是Apache可以同时处理的请求,是对Apache性能影响最大的参数。其缺省值150是远远不够的,如果请求总数已达到这个值(可通过ps -ef|grep http|wc -l来确认),那么后面的请求就要排队,直到某个已处理请求完毕。这就是系统资源还剩下很多而HTTP访问却很慢的主要原因。虽然理论上这个值越大,可以处理的请求就越多,但Apache默认的限制不能大于256。ServerLimit指令无须重编译Apache就可以加大MaxClients。ServerLimt应该放在第一个位置,放在其他指令之间不起作用(不明白原因)。
代码如下
<IfModule prefork.c>
ServerLimit 10000
StartServers 5
MinSpareServers 5
MaxSpareServers 10
MaxClients 10000
MaxRequestsPerChild 0
</IfModule>
Worker
相对于prefork,worker全新的支持多线程和多进程混合模型的MPM。由于使用线程来处理,所以可以处理相对海量的请求,而系统资源的开销要小于基于进程的服务器。但是,worker也使用了多进程,每个进程又生成多个线程,以获得基于进程服务器的稳定性。在configure --with-mpm=worker后,进行make编译、make install安装。在缺省生成的httpd-mpm.conf中有以下默认配置段:
代码如下
<IfModule worker.c>
StartServers 2
MaxClients 150
MinSpareThreads 25
MaxSpareThreads 75
ThreadsPerChild 25
MaxRequestsPerChild 0
</IfModule>
Worker由主控制进程生成“StartServers”个子进程,每个子进程中包含固定的ThreadsPerChild线程数,各个线程独立地处理请求。同样,为了不在请求到来时再生成线程,MinSpareThreads和MaxSpareThreads设置了最少和最多的空闲线程数;而MaxClients设置了同时连入的clients最大总数。如果现有子进程中的线程总数不能满足负载,控制进程将派生新的子进程。MinSpareThreads和MaxSpareThreads的最大缺省值分别是75和250。这两个参数对Apache的性能影响并不大,可以按照实际情况相应调节。ThreadsPerChild是worker MPM中与性能相关最密切的指令。ThreadsPerChild的最大缺省值是64,如果负载较大,64也是不够的。这时要显式使用ThreadLimit指令,它的最大缺省值是20000。Worker模式下所能同时处理的请求总数是由子进程总数乘以ThreadsPerChild值决定的,应该大于等于MaxClients。如果负载很大,现有的子进程数不能满足时,控制进程会派生新的子进程。默认最大的子进程总数是16,加大时也需要显式声明ServerLimit(最大值是20000)。需要注意的是,如果显式声明了ServerLimit,那么它乘以ThreadsPerChild的值必须大于等于MaxClients,而且MaxClients必须是ThreadsPerChild的整数倍,否则Apache将会自动调节到一个相应值。
代码如下
<IfModule worker.c>
ServerLimit 25
ThreadLimit 200
StartServers 3
MaxClients 2000
MinSpareThreads 50
MaxSpareThreads 200
ThreadsPerChild 100
MaxRequestsPerChild 0
</IfModule>
下面是利用Apache自带的测试工具ab对Server进行测试的情况(设定请求的index页面为6bytes,Apache Server配置2cpu 2G memory),cpu%为cpu占用率,mem为内存使用量(M为单位),RequestsPerSecond为每秒处理的请求数。
1、Prefor方式
(ServerLimit,StartServer,MinSpareServers,MaxSpareServers,MaxClients,MaxRequestPerChild)
-n/-c(ab参数) Cpu% Mem Requestspersecond
(-,5,5,10,150,0)
100000/100 28.8 285 8434
100000/200 29.2 304 8032
100000/500 25.3 323 7348
100000/1000 24.4 330 5886
(10000,5,5,10,500,0)
100000/100 28.7 371 8345
100000/200 27.4 389 7929
100000/500 24.9 417 7229
100000/1000 23.4 437 6676
(10000,5,5,10,1000,0)
100000/100 28.8 408 8517
100000/200 27.0 422 8045
100000/500 24.2 455 7236
100000/1000 22.5 470 6570
(10000,5,5,10,1500,0)
100000/100 29.6 330 8407
100000/200 28.1 349 8014
100000/500 26.4 380 7290
100000/1000 24.0 400 6686
2、Worker方式
(ServerLimt,Threadlimt,Startservers,MaxClients,MinspareThread,MaxspareThread,ThreadperChild,MaxRequestPerChild)
-n/-c(ab参数) cpu% mem RequestsperSecond
(50,500,5,10000,50,200,200,0)
100000/100 18.6 188 6020
100000/200 20.1 195 5892
100000/500 19.8 209 5708
100000/1000 22.2 218 6081
(100,500,5,10000,50,200,100,0)
100000/100 24.5 240 6919
100000/200 23.6 247 6798
100000/500 24.6 254 6827
100000/1000 22.3 271 6114
(200,500,5,10000,50,200,50,0)
100000/100 27.3 301 7781
100000/200 27.4 307 7789
100000/500 26.0 320 7141
100000/1000 21.8 344 6110
相对来说,prefork方式速度要稍高于worker,然而它需要的cpu和memory资源也稍多于woker。
下面贴出我的系统mpm文件
代码如下
#
# Server-Pool Management (MPM specific)
#
#
# PidFile: The file in which the server should record its process
# identification number when it starts.
#
# Note that this is the default PidFile for most MPMs.
#
<IfModule !mpm_netware_module>
PidFile "logs/httpd.pid"
</IfModule>
#
# The accept serialization lock file MUST BE STORED ON A LOCAL DISK.
#
<IfModule !mpm_winnt_module>
<IfModule !mpm_netware_module>
LockFile "logs/accept.lock"
</IfModule>
</IfModule>
#
# Only one of the below sections will be relevant on your
# installed httpd. Use "apachectl -l" to find out the
# active mpm.
#
# prefork MPM
# StartServers: number of server processes to start
# MinSpareServers: minimum number of server processes which are kept spare
# MaxSpareServers: maximum number of server processes which are kept spare
# MaxClients: maximum number of server processes allowed to start
# MaxRequestsPerChild: maximum number of requests a server process serves
<IfModule mpm_prefork_module>
StartServers 5
MinSpareServers 5
MaxSpareServers 10
MaxClients 150
MaxRequestsPerChild 0
</IfModule>
# worker MPM
# StartServers: initial number of server processes to start
# MaxClients: maximum number of simultaneous client connections
# MinSpareThreads: minimum number of worker threads which are kept spare
# MaxSpareThreads: maximum number of worker threads which are kept spare
# ThreadsPerChild: constant number of worker threads in each server process
# MaxRequestsPerChild: maximum number of requests a server process serves
<IfModule mpm_worker_module>
StartServers 2
MaxClients 150
MinSpareThreads 25
MaxSpareThreads 75
ThreadsPerChild 25
MaxRequestsPerChild 0
</IfModule>
# BeOS MPM
# StartThreads: how many threads do we initially spawn?
# MaxClients: max number of threads we can have (1 thread == 1 client)
# MaxRequestsPerThread: maximum number of requests each thread will process
<IfModule mpm_beos_module>
StartThreads 10
MaxClients 50
MaxRequestsPerThread 10000
</IfModule>
# NetWare MPM
# ThreadStackSize: Stack size allocated for each worker thread
# StartThreads: Number of worker threads launched at server startup
# MinSpareThreads: Minimum number of idle threads, to handle request spikes
# MaxSpareThreads: Maximum number of idle threads
# MaxThreads: Maximum number of worker threads alive at the same time
# MaxRequestsPerChild: Maximum number of requests a thread serves. It is
# recommended that the default value of 0 be set for this
# directive on NetWare. This will allow the thread to
# continue to service requests indefinitely.
<IfModule mpm_netware_module>
ThreadStackSize 65536
StartThreads 250
MinSpareThreads 25
MaxSpareThreads 250
MaxThreads 1000
MaxRequestsPerChild 0
MaxMemFree 100
</IfModule>
# OS/2 MPM
# StartServers: Number of server processes to maintain
# MinSpareThreads: Minimum number of idle threads per process,
# to handle request spikes
# MaxSpareThreads: Maximum number of idle threads per process
# MaxRequestsPerChild: Maximum number of connections per server process
<IfModule mpm_mpmt_os2_module>
StartServers 2
MinSpareThreads 5
MaxSpareThreads 10
MaxRequestsPerChild 0
</IfModule>
# WinNT MPM
# ThreadsPerChild: constant number of worker threads in the server process
# MaxRequestsPerChild: maximum number of requests a server process serves
<IfModule mpm_winnt_module>
ThreadsPerChild 150
MaxRequestsPerChild 0
</IfModule>