Linux操作系统中关于负载的定义

向右看齐 2022-02-04 06:11 156阅读 0赞

<table style="font-family:'Helvetica Neue', Helvetica, 'PingFang SC', 'Hiragino Sans GB', 'Microsoft YaHei', '微软雅黑', Arial, sans-serif;font-size:16px;border-collapse:collapse;color:rgb(51,51,51);"> 
 <tbody> 
 <tr> 
 <td style="background:#FFFFFF;border:1px solid rgb(221,221,221);"> 
 <div> 
 
 <div style="text-indent:2em;">
 使用uptime或者top命令，都可以看到一个负载的输出，形如load average: 0.00, 0.03, 0.00，这个负载到底是什么东西呢，man文档里只是一笔带过，没有具体的给出负载的定义。 
 
 负载的统计，必然是由内核完成的，因此在内核源码中找答案是再好不过的事情了，找来2.6.21的内核源码，开始探索。 
 
 节选部分源码： 
 
 </div> 
 <table style="border-collapse:collapse;"> 
 <tbody> 
 <tr> 
 <td style="background:#FFFFFF;border:1px solid rgb(221,221,221);"> 
 <div style="line-height:24px;">
 //kernel/timer.c 
 
 <code style="font-family:'Helvetica Neue', Helvetica, 'PingFang SC', 'Hiragino Sans GB', 'Microsoft YaHei', '微软雅黑', Arial, sans-serif;"></code> 
 </div> 
 <div style="line-height:24px;"> 
 <code style="font-family:'Helvetica Neue', Helvetica, 'PingFang SC', 'Hiragino Sans GB', 'Microsoft YaHei', '微软雅黑', Arial, sans-serif;">1254 active_tasks&nbsp;=&nbsp;count_active_tasks(); 1256 CALC_LOAD(avenrun[0],&nbsp;EXP_1,&nbsp;active_tasks); 1257 CALC_LOAD(avenrun[1],&nbsp;EXP_5,&nbsp;active_tasks); 1258 CALC_LOAD(avenrun[2],&nbsp;EXP_15,&nbsp;active_tasks);</code> 
 </div> 
 <div style="line-height:24px;"> 
 
 <code style="font-family:'Helvetica Neue', Helvetica, 'PingFang SC', 'Hiragino Sans GB', 'Microsoft YaHei', '微软雅黑', Arial, sans-serif;"></code> 
 </div> 
 <div style="line-height:24px;"> 
 <code style="font-family:'Helvetica Neue', Helvetica, 'PingFang SC', 'Hiragino Sans GB', 'Microsoft YaHei', '微软雅黑', Arial, sans-serif;">//include/linux/sched.h</code> 
 </div> 
 <div style="line-height:24px;"> 
 <code style="font-family:'Helvetica Neue', Helvetica, 'PingFang SC', 'Hiragino Sans GB', 'Microsoft YaHei', '微软雅黑', Arial, sans-serif;">&nbsp;110 #define FSHIFT&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; 11&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; /* nr of bits of precision */ &nbsp;111 #define FIXED_1&nbsp;&nbsp;&nbsp;&nbsp; (1&lt; &nbsp;112 #define LOAD_FREQ&nbsp;&nbsp; (5*HZ)&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; /* 5 sec intervals */ &nbsp;113 #define EXP_1&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; 1884&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; /* 1/exp(5sec/1min) as fixed-point */ &nbsp;114 #define EXP_5&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; 2014&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; /* 1/exp(5sec/5min) */ &nbsp;115 #define EXP_15&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; 2037&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; /* 1/exp(5sec/15min) */ &nbsp;117 #define CALC_LOAD(load,exp,n) \ &nbsp;118&nbsp;&nbsp;&nbsp;&nbsp; load *= exp; \ &nbsp;119&nbsp;&nbsp;&nbsp;&nbsp; load += n*(FIXED_1-exp); \ &nbsp;120&nbsp;&nbsp;&nbsp;&nbsp; load &gt;&gt;= FSHIFT;&nbsp; )&gt;</code> 
 </div> </td> 
 </tr> 
 </tbody> 
 </table> 
 <div> 
 load(t) = ( load(t-1)*exp(i) + n(t)*(2048-exp(i)) ) / 2048 
 </div> 
 <div>
 　　load(t-1)为上次计算出的结果 
 </div> 
 <div>
 　　n(t)为t时刻的活动进程数 
 </div> 
 <div>
 　　计算方式是累加各个CPU的运行队列中running和uninterruptible的值　再乘以2048 
 </div> 
 <div>
 　　计算方式如下： 
 
 </div> 
 <table style="border-collapse:collapse;"> 
 <tbody> 
 <tr> 
 <td style="background:#FFFFFF;border:1px solid rgb(221,221,221);"> 
 <div style="line-height:24px;"> 
 <code style="font-family:'Helvetica Neue', Helvetica, 'PingFang SC', 'Hiragino Sans GB', 'Microsoft YaHei', '微软雅黑', Arial, sans-serif;">1946&nbsp;unsigned&nbsp;long&nbsp;nr_active(void) 1947&nbsp;{ 1948&nbsp;unsigned&nbsp;long&nbsp;i,&nbsp;running&nbsp;=&nbsp;0,&nbsp;uninterruptible&nbsp;=&nbsp;0; 1949 1950 for_each_online_cpu(i)&nbsp;{ 1951 running&nbsp;+=&nbsp;cpu_rq(i)-&gt;nr_running; 1952 uninterruptible&nbsp;+=&nbsp;cpu_rq(i)-&gt;nr_uninterruptible; 1953&nbsp;} 1954 1955&nbsp;if&nbsp;(unlikely((long)uninterruptible&nbsp;&lt;&nbsp;0)) 1956 uninterruptible&nbsp;=&nbsp;0; 1957 1958&nbsp;return&nbsp;running&nbsp;+&nbsp;uninterruptible; 1959&nbsp;} 1226&nbsp;static&nbsp;unsigned&nbsp;long&nbsp;count_active_tasks(void) 1227&nbsp;{ 1228&nbsp;return&nbsp;nr_active()&nbsp;*&nbsp;FIXED_1; 1229&nbsp;}</code> 
 </div> </td> 
 </tr> 
 </tbody> 
 </table> 
 <div> 
 exp(1) = 1884 
 exp(5) = 2014 
 exp(15) = 2037 
 exp(i) = 2048 * e^(-1/12/i) 
 
 从本质上看负载是完全由过去的一段时间里每个CPU上的活动进程数决定的，但并不是在数值上等同于每秒钟需要进行调度的进程数，具体的计算过程是个比较复杂的过程。 
 </div> 
 </div> </td> 
 </tr> 
 </tbody> 
</table>

本文转自 liang831002 51CTO博客，原文链接：http://blog.51cto.com/leo0216/70834，如需转载请自行联系原作者