1*4882a593Smuzhiyun======== 2*4882a593SmuzhiyunCPU 负载 3*4882a593Smuzhiyun======== 4*4882a593Smuzhiyun 5*4882a593SmuzhiyunLinux通过``/proc/stat``和``/proc/uptime``导出各种信息，用户空间工具 6*4882a593Smuzhiyun如top(1)使用这些信息计算系统花费在某个特定状态的平均时间。 7*4882a593Smuzhiyun例如： 8*4882a593Smuzhiyun 9*4882a593Smuzhiyun $ iostat 10*4882a593Smuzhiyun Linux 2.6.18.3-exp (linmac) 02/20/2007 11*4882a593Smuzhiyun 12*4882a593Smuzhiyun avg-cpu: %user %nice %system %iowait %steal %idle 13*4882a593Smuzhiyun 10.01 0.00 2.92 5.44 0.00 81.63 14*4882a593Smuzhiyun 15*4882a593Smuzhiyun ... 16*4882a593Smuzhiyun 17*4882a593Smuzhiyun这里系统认为在默认采样周期內有10.01%的时间工作在用户空间，2.92%的时 18*4882a593Smuzhiyun间用在系统空间，总体上有81.63%的时间是空闲的。 19*4882a593Smuzhiyun 20*4882a593Smuzhiyun大多数情况下``/proc/stat``的信息几乎真实反映了系统信息，然而，由于内 21*4882a593Smuzhiyun核采集这些数据的方式/时间的特点，有时这些信息根本不可靠。 22*4882a593Smuzhiyun 23*4882a593Smuzhiyun那么这些信息是如何被搜集的呢？每当时间中断触发时，内核查看此刻运行的 24*4882a593Smuzhiyun进程类型，并增加与此类型/状态进程对应的计数器的值。这种方法的问题是 25*4882a593Smuzhiyun在两次时间中断之间系统（进程）能够在多种状态之间切换多次，而计数器只 26*4882a593Smuzhiyun增加最后一种状态下的计数。 27*4882a593Smuzhiyun 28*4882a593Smuzhiyun举例 29*4882a593Smuzhiyun--- 30*4882a593Smuzhiyun 31*4882a593Smuzhiyun假设系统有一个进程以如下方式周期性地占用cpu:: 32*4882a593Smuzhiyun 33*4882a593Smuzhiyun 两个时钟中断之间的时间线 34*4882a593Smuzhiyun |-----------------------| 35*4882a593Smuzhiyun ^ ^ 36*4882a593Smuzhiyun |_ 开始运行 | 37*4882a593Smuzhiyun |_ 开始睡眠 38*4882a593Smuzhiyun （很快会被唤醒） 39*4882a593Smuzhiyun 40*4882a593Smuzhiyun在上面的情况下，根据``/proc/stat``的信息（由于当系统处于空闲状态时， 41*4882a593Smuzhiyun时间中断经常会发生）系统的负载将会是0 42*4882a593Smuzhiyun 43*4882a593Smuzhiyun大家能够想象内核的这种行为会发生在许多情况下，这将导致``/proc/stat`` 44*4882a593Smuzhiyun中存在相当古怪的信息:: 45*4882a593Smuzhiyun 46*4882a593Smuzhiyun /* gcc -o hog smallhog.c */ 47*4882a593Smuzhiyun #include <time.h> 48*4882a593Smuzhiyun #include <limits.h> 49*4882a593Smuzhiyun #include <signal.h> 50*4882a593Smuzhiyun #include <sys/time.h> 51*4882a593Smuzhiyun #define HIST 10 52*4882a593Smuzhiyun 53*4882a593Smuzhiyun static volatile sig_atomic_t stop; 54*4882a593Smuzhiyun 55*4882a593Smuzhiyun static void sighandler (int signr) 56*4882a593Smuzhiyun { 57*4882a593Smuzhiyun (void) signr; 58*4882a593Smuzhiyun stop = 1; 59*4882a593Smuzhiyun } 60*4882a593Smuzhiyun static unsigned long hog (unsigned long niters) 61*4882a593Smuzhiyun { 62*4882a593Smuzhiyun stop = 0; 63*4882a593Smuzhiyun while (!stop && --niters); 64*4882a593Smuzhiyun return niters; 65*4882a593Smuzhiyun } 66*4882a593Smuzhiyun int main (void) 67*4882a593Smuzhiyun { 68*4882a593Smuzhiyun int i; 69*4882a593Smuzhiyun struct itimerval it = { .it_interval = { .tv_sec = 0, .tv_usec = 1 }, 70*4882a593Smuzhiyun .it_value = { .tv_sec = 0, .tv_usec = 1 } }; 71*4882a593Smuzhiyun sigset_t set; 72*4882a593Smuzhiyun unsigned long v[HIST]; 73*4882a593Smuzhiyun double tmp = 0.0; 74*4882a593Smuzhiyun unsigned long n; 75*4882a593Smuzhiyun signal (SIGALRM, &sighandler); 76*4882a593Smuzhiyun setitimer (ITIMER_REAL, &it, NULL); 77*4882a593Smuzhiyun 78*4882a593Smuzhiyun hog (ULONG_MAX); 79*4882a593Smuzhiyun for (i = 0; i < HIST; ++i) v[i] = ULONG_MAX - hog (ULONG_MAX); 80*4882a593Smuzhiyun for (i = 0; i < HIST; ++i) tmp += v[i]; 81*4882a593Smuzhiyun tmp /= HIST; 82*4882a593Smuzhiyun n = tmp - (tmp / 3.0); 83*4882a593Smuzhiyun 84*4882a593Smuzhiyun sigemptyset (&set); 85*4882a593Smuzhiyun sigaddset (&set, SIGALRM); 86*4882a593Smuzhiyun 87*4882a593Smuzhiyun for (;;) { 88*4882a593Smuzhiyun hog (n); 89*4882a593Smuzhiyun sigwait (&set, &i); 90*4882a593Smuzhiyun } 91*4882a593Smuzhiyun return 0; 92*4882a593Smuzhiyun } 93*4882a593Smuzhiyun 94*4882a593Smuzhiyun 95*4882a593Smuzhiyun参考 96*4882a593Smuzhiyun--- 97*4882a593Smuzhiyun 98*4882a593Smuzhiyun- http://lkml.org/lkml/2007/2/12/6 99*4882a593Smuzhiyun- Documentation/filesystems/proc.rst (1.8) 100*4882a593Smuzhiyun 101*4882a593Smuzhiyun 102*4882a593Smuzhiyun谢谢 103*4882a593Smuzhiyun--- 104*4882a593Smuzhiyun 105*4882a593SmuzhiyunCon Kolivas, Pavel Machek 106