Linux中CPU亲和性(affinity)
0、准备知识
超线程技术(Hyper-Threading):就是利用特殊的硬件指令,把两个逻辑内核(CPU core)模拟成两个物理芯片,
让单个处理器都能使用线程级并行计算,进而兼容多线程操作系统和软件,减少了CPU的闲置时间,提高的CPU的运行效率。
我们常听到的双核四线程/四核八线程指的就是支持超线程技术的CPU.
物理CPU:机器上安装的实际CPU, 比如说你的主板上安装了一个8核CPU,那么物理CPU个数就是1个,所以物理CPU个数就是主板上安装的CPU个数。
逻辑CPU:一般情况,我们认为一颗CPU可以有多核,加上intel的超线程技术(HT), 可以在逻辑上再分一倍数量的CPU core出来;
逻辑CPU数量 = 物理CPU数量 x CPU cores x 2(如果支持并开启HT) //前提是CPU的型号一致,如果不一致只能一个一个的加起来,不用直接乘以物理CPU数量 //比如你的电脑安装了一块4核CPU,并且支持且开启了超线程(HT)技术,那么逻辑CPU数量 = 1 × 4 × 2 = 8
Linux下查看CPU相关信息, CPU的信息主要都在/proc/cupinfo中,
# 查看物理CPU个数 cat /proc/cpuinfo|grep "physical id"|sort -u|wc -l # 查看每个物理CPU中core的个数(即核数) cat /proc/cpuinfo|grep "cpu cores"|uniq # 查看逻辑CPU的个数 cat /proc/cpuinfo|grep "processor"|wc -l # 查看CPU的名称型号 cat /proc/cpuinfo|grep "name"|cut -f2 -d:|uniq
Linux查看某个进程运行在哪个逻辑CPU上
ps -eo pid,args,psr
#参数的含义: pid - 进程ID args - 该进程执行时传入的命令行参数 psr - 分配给进程的逻辑CPU
例子:
[~]# ps -eo pid,args,psr | grep nginx
9073 nginx: master process /usr/ 1
9074 nginx: worker process 0
9075 nginx: worker process 1
9076 nginx: worker process 2
9077 nginx: worker process 3
13857 grep nginx 3
Linux查看线程的TID
TID就是Thread ID,他和POSIX中pthread_t表示的线程ID完全不是同一个东西.
Linux中的POSIX线程库实现的线程其实也是一个轻量级进程(LWP),这个TID就是这个线程的真实PID.
但是又不能通过getpid()函数获取,Linux中定义了gettid()这个接口,但是通常都是未实现的,所以需要使用下面的方式获取TID。
//program #include <sys/syscall.h> pid_t tid; tid = syscall(__NR_gettid);// or syscall(SYS_gettid) //command-line 3种方法(推荐第三种方法) (1)ps -efL | grep prog_name (2)ls /proc/pid/task //文件夹名即TID
(3)ps -To 'pid,lwp,psr,cmd' -p PID
1、CPU亲和性(亲和力)
1.1 基本概念
CPU affinity 是一种调度属性(scheduler property), 它可以将一个进程"绑定" 到一个或一组CPU上.
在SMP(Symmetric Multi-Processing对称多处理)架构下,Linux调度器(scheduler)会根据CPU affinity的设置让指定的进程运行在"绑定"的CPU上,而不会在别的CPU上运行.
Linux调度器同样支持自然CPU亲和性(natural CPU affinity): 调度器会试图保持进程在相同的CPU上运行, 这意味着进程通常不会在处理器之间频繁迁移,进程迁移的频率小就意味着产生的负载小。
因为程序的作者比调度器更了解程序,所以我们可以手动地为其分配CPU核,而不会过多地占用CPU0,或是让我们关键进程和一堆别的进程挤在一起,所有设置CPU亲和性可以使某些程序提高性能。
1.2 表示方法
CPU affinity 使用位掩码(bitmask)表示, 每一位都表示一个CPU, 置1表示"绑定".
最低位表示第一个逻辑CPU, 最高位表示最后一个逻辑CPU.
CPU affinity典型的表示方法是使用16进制,具体如下.
0x00000001 is processor #0 0x00000003 is processors #0 and #1 0xFFFFFFFF is all processors (#0 through #31)
2、taskset命令
taskset命名用于获取或者设定CPU亲和性.
# 命令行形式 taskset [options] mask command [arg]...
taskset [options] -p [mask] pid
PARAMETER
mask : cpu亲和性,当没有-c选项时, 其值前无论有没有0x标记都是16进制的,
当有-c选项时,其值是十进制的.
command : 命令或者可执行程序
arg : command的参数
pid : 进程ID,可以通过ps/top/pidof等命令获取
OPTIONS
-a, --all-tasks (旧版本中没有这个选项)
这个选项涉及到了linux中TID的概念,他会将一个进程中所有的TID都执行一次CPU亲和性设置.
TID就是Thread ID,他和POSIX中pthread_t表示的线程ID完全不是同一个东西.
Linux中的POSIX线程库实现的线程其实也是一个进程(LWP),这个TID就是这个线程的真实PID. -p, --pid 操作已存在的PID,而不是加载一个新的程序 -c, --cpu-list 声明CPU的亲和力使用数字表示而不是用位掩码表示. 例如 0,5,7,9-11. -h, --help display usage information and exit -V, --version output version information and exit
USAGE
1) 使用指定的CPU亲和性运行一个新程序
taskset [-c] mask command [arg]...
举例:使用CPU0运行ls命令显示/etc/init.d下的所有内容
taskset -c 0 ls -al /etc/init.d/
2) 显示已经运行的进程的CPU亲和性
taskset -p pid
举例:查看init进程(PID=1)的CPU亲和性
taskset -p 1
3) 改变已经运行进程的CPU亲和力
taskset -p[c] mask pid
举例:打开2个终端,在第一个终端运行top命令,第二个终端中
首先运行:[~]# ps -eo pid,args,psr | grep top #获取top命令的pid和其所运行的CPU号
其次运行:[~]# taskset -cp 新的CPU号 pid #更改top命令运行的CPU号
最后运行:[~]# ps -eo pid,args,psr | grep top #查看是否更改成功
PERMISSIONS
一个用户要设定一个进程的CPU亲和性,如果目标进程是该用户的,则可以设置,如果是其他用户的,则会设置失败,提示 Operation not permitted.当然root用户没有任何限制.
任何用户都可以获取任意一个进程的CPU亲和性.
taskset命令其实就是使用sched_getaffinity()和sched_setaffinity()接口实现的,相信看完了第3节你也能自己实现一个taskset命令.
有兴趣的可以看一下其源代码:ftp://ftp.kernel.org/pub/linux/utils/util-linux/vX.YZ/util-linux-X.YZ-xxx.tar.gz /schedutils/taskset.c
3、编程API
下面是用用于设置和获取CPU亲和性相关的API.
#define _GNU_SOURCE #include <sched.h> #include <pthread.h> //for pthread functions(last 4) 注意<pthread.h>包含<sched.h> /* MACRO */ /* The following macros are provided to operate on the CPU set set */ /* Clears set, so that it contains no CPUs */ void CPU_ZERO(cpu_set_t *set); void CPU_ZERO_S(size_t setsize, cpu_set_t *set); /* Add CPU cpu to set */ void CPU_SET(int cpu, cpu_set_t *set); void CPU_SET_S(int cpu, size_t setsize, cpu_set_t *set); /* Remove CPU cpu from set */ void CPU_CLR(int cpu, cpu_set_t *set); void CPU_CLR_S(int cpu, size_t setsize, cpu_set_t *set); /* Test to see if CPU cpu is a member of set */ int CPU_ISSET(int cpu, cpu_set_t *set); int CPU_ISSET_S(int cpu, size_t setsize, cpu_set_t *set); /* Return the number of CPUs in set */ void CPU_COUNT(cpu_set_t *set); void CPU_COUNT_S(size_t setsize, cpu_set_t *set); /* The following macros perform logical operations on CPU sets */ /* Store the logical AND of the sets srcset1 and srcset2 in destset (which may be one of the source sets). */ void CPU_AND(cpu_set_t *destset, cpu_set_t *srcset1, cpu_set_t *srcset2); void CPU_AND_S(size_t setsize, cpu_set_t *destset, cpu_set_t *srcset1, cpu_set_t *srcset2); /* Store the logical OR of the sets srcset1 and srcset2 in destset (which may be one of the source sets). */ void CPU_OR(cpu_set_t *destset, cpu_set_t *srcset1, cpu_set_t *srcset2); void CPU_OR_S(size_t setsize, cpu_set_t *destset, cpu_set_t *srcset1, cpu_set_t *srcset2); /* Store the logical XOR of the sets srcset1 and srcset2 in destset (which may be one of the source sets). */ void CPU_XOR(cpu_set_t *destset, cpu_set_t *srcset1, cpu_set_t *srcset2); void CPU_XOR_S(size_t setsize, cpu_set_t *destset, cpu_set_t *srcset1, cpu_set_t *srcset2); /* Test whether two CPU set contain exactly the same CPUs. */ int CPU_EQUAL(cpu_set_t *set1, cpu_set_t *set2); int CPU_EQUAL_S(size_t setsize, cpu_set_t *set1, cpu_set_t *set2); /* The following macros are used to allocate and deallocate CPU sets: */ /* Allocate a CPU set large enough to hold CPUs in the range 0 to num_cpus-1 */ cpu_set_t *CPU_ALLOC(int num_cpus); /* Return the size in bytes of the CPU set that would be needed to hold CPUs in the range 0 to num_cpus-1. This macro provides the value that can be used for the setsize argument in the CPU_*_S() macros */ size_t CPU_ALLOC_SIZE(int num_cpus); /* Free a CPU set previously allocated by CPU_ALLOC(). */ void CPU_FREE(cpu_set_t *set); /* API */ /* Set the CPU affinity for a task */ int sched_setaffinity(pid_t pid, size_t cpusetsize, cpu_set_t *mask); /* Get the CPU affinity for a task */ int sched_getaffinity(pid_t pid, size_t cpusetsize, cpu_set_t *mask); /* set CPU affinity attribute in thread attributes object */ int pthread_attr_setaffinity_np(pthread_attr_t *attr, size_t cpusetsize, const cpu_set_t *cpuset); /* get CPU affinity attribute in thread attributes object */ int pthread_attr_getaffinity_np(const pthread_attr_t *attr, size_t cpusetsize, cpu_set_t *cpuset); /* set CPU affinity of a thread */ int pthread_setaffinity_np(pthread_t thread, size_t cpusetsize, const cpu_set_t *cpuset); /* get CPU affinity of a thread */ int pthread_getaffinity_np(pthread_t thread, size_t cpusetsize, cpu_set_t *cpuset);
相关的宏通常都分为2种,一种是带_S后缀的,一种不是不带_S后缀的, 从声明上看带_S后缀的宏都多出一个参数 setsize.
从功能上看他们的区别是带_S后缀的宏是用于操作动态申请的CPU set(s),所谓的动态申请其实就是使用宏 CPU_ALLOC 申请,
参数setsize 可以是通过宏 CPU_ALLOC_SIZE 获得,两者的用法详见下面的例子.
相关的API只有6个, 前2个是用来设置进程的CPU亲和性,需要注意的一点是,当这2个API的第一个参数pid为0时,表示使用调用进程的进程ID;
后4个是用来设置线程的CPU亲和性。其实sched_setaffinity()也可以用来设置线程的CPU的亲和性,也就是taskset “-a”选项中提到的TID概念。
3.1 例子一:使用2种方式(带和不带_S后缀的宏)获取当前进程的CPU亲和性
#define _GNU_SOURCE #include <sched.h> #include <unistd.h> /* sysconf */ #include <stdlib.h> /* exit */ #include <stdio.h> int main(void) { int i, nrcpus; cpu_set_t mask; unsigned long bitmask = 0; CPU_ZERO(&mask); /* Get the CPU affinity for a pid */ if (sched_getaffinity(0, sizeof(cpu_set_t), &mask) == -1) { perror("sched_getaffinity"); exit(EXIT_FAILURE); } /* get logical cpu number */ nrcpus = sysconf(_SC_NPROCESSORS_CONF); for (i = 0; i < nrcpus; i++) { if (CPU_ISSET(i, &mask)) { bitmask |= (unsigned long)0x01 << i; printf("processor #%d is set\n", i); } } printf("bitmask = %#lx\n", bitmask); exit(EXIT_SUCCESS); } /*----------------------------------------------------------------*/ #define _GNU_SOURCE #include <sched.h> #include <unistd.h> /* sysconf */ #include <stdlib.h> /* exit */ #include <stdio.h> int main(void) { int i, nrcpus; cpu_set_t *pmask; size_t cpusize; unsigned long bitmask = 0; /* get logical cpu number */ nrcpus = sysconf(_SC_NPROCESSORS_CONF); pmask = CPU_ALLOC(nrcpus); cpusize = CPU_ALLOC_SIZE(nrcpus); CPU_ZERO_S(cpusize, pmask); /* Get the CPU affinity for a pid */ if (sched_getaffinity(0, cpusize, pmask) == -1) { perror("sched_getaffinity"); CPU_FREE(pmask); exit(EXIT_FAILURE); } for (i = 0; i < nrcpus; i++) { if (CPU_ISSET_S(i, cpusize, pmask)) { bitmask |= (unsigned long)0x01 << i; printf("processor #%d is set\n", i); } } printf("bitmask = %#lx\n", bitmask); CPU_FREE(pmask); exit(EXIT_SUCCESS); }
执行结果如下(4核CPU):
[cpu_affinity #1]$ gcc -g -Wall cpu_affinity.c [cpu_affinity #2]$ taskset 1 ./a.out processor #0 is set bitmask = 0x1 [cpu_affinity #3]$ taskset 1 ./a.out processor #0 is set bitmask = 0x1 [cpu_affinity #4]$ taskset 2 ./a.out processor #1 is set bitmask = 0x2 [cpu_affinity #5]$ taskset 3 ./a.out processor #0 is set processor #1 is set bitmask = 0x3 [cpu_affinity #6]$ taskset 4 ./a.out processor #2 is set bitmask = 0x4 [cpu_affinity #7]$ taskset 5 ./a.out processor #0 is set processor #2 is set bitmask = 0x5 [cpu_affinity #8]$ taskset 6 ./a.out processor #1 is set processor #2 is set bitmask = 0x6 [cpu_affinity #9]$ taskset 7 ./a.out processor #0 is set processor #1 is set processor #2 is set bitmask = 0x7 [cpu_affinity #10]$ taskset 8 ./a.out processor #3 is set bitmask = 0x8 [cpu_affinity #11]$ taskset 9 ./a.out processor #0 is set processor #3 is set bitmask = 0x9 [cpu_affinity #12]$ taskset A ./a.out processor #1 is set processor #3 is set bitmask = 0xa [cpu_affinity #13]$ taskset B ./a.out processor #0 is set processor #1 is set processor #3 is set bitmask = 0xb [cpu_affinity #14]$ taskset C ./a.out processor #2 is set processor #3 is set bitmask = 0xc [cpu_affinity #15]$ taskset D ./a.out processor #0 is set processor #2 is set processor #3 is set bitmask = 0xd [cpu_affinity #16]$ taskset E ./a.out processor #1 is set processor #2 is set processor #3 is set bitmask = 0xe [cpu_affinity #17]$ taskset F ./a.out processor #0 is set processor #1 is set processor #2 is set processor #3 is set bitmask = 0xf [cpu_affinity #18]$ taskset 0 ./a.out sched_setaffinity: Invalid argument failed to set pid 0's affinity.
3.2 例子二:设置进程的CPU亲和性后再获取显示CPU亲和性
#define _GNU_SOURCE #include <sched.h> #include <unistd.h> /* sysconf */ #include <stdlib.h> /* exit */ #include <stdio.h> int main(void) { int i, nrcpus; cpu_set_t mask; unsigned long bitmask = 0; CPU_ZERO(&mask); CPU_SET(0, &mask); /* add CPU0 to cpu set */ CPU_SET(2, &mask); /* add CPU2 to cpu set */ /* Set the CPU affinity for a pid */ if (sched_setaffinity(0, sizeof(cpu_set_t), &mask) == -1) { perror("sched_setaffinity"); exit(EXIT_FAILURE); } CPU_ZERO(&mask); /* Get the CPU affinity for a pid */ if (sched_getaffinity(0, sizeof(cpu_set_t), &mask) == -1) { perror("sched_getaffinity"); exit(EXIT_FAILURE); } /* get logical cpu number */ nrcpus = sysconf(_SC_NPROCESSORS_CONF); for (i = 0; i < nrcpus; i++) { if (CPU_ISSET(i, &mask)) { bitmask |= (unsigned long)0x01 << i; printf("processor #%d is set\n", i); } } printf("bitmask = %#lx\n", bitmask); exit(EXIT_SUCCESS); }
3.3 例子三:设置线程的CPU属性后再获取显示CPU亲和性
这个例子来源于Linux的man page.
#define _GNU_SOURCE #include <pthread.h> //不用再包含<sched.h> #include <stdio.h> #include <stdlib.h> #include <errno.h> #define handle_error_en(en, msg) \ do { errno = en; perror(msg); exit(EXIT_FAILURE); } while (0) int main(int argc, char *argv[]) { int s, j; cpu_set_t cpuset; pthread_t thread; thread = pthread_self(); /* Set affinity mask to include CPUs 0 to 7 */ CPU_ZERO(&cpuset); for (j = 0; j < 8; j++) CPU_SET(j, &cpuset); s = pthread_setaffinity_np(thread, sizeof(cpu_set_t), &cpuset); if (s != 0) { handle_error_en(s, "pthread_setaffinity_np"); } /* Check the actual affinity mask assigned to the thread */ s = pthread_getaffinity_np(thread, sizeof(cpu_set_t), &cpuset); if (s != 0) { handle_error_en(s, "pthread_getaffinity_np"); } printf("Set returned by pthread_getaffinity_np() contained:\n"); for (j = 0; j < CPU_SETSIZE; j++) //CPU_SETSIZE 是定义在<sched.h>中的宏,通常是1024 { if (CPU_ISSET(j, &cpuset)) { printf(" CPU %d\n", j); } } exit(EXIT_SUCCESS); }
3.4 例子四:使用seched_setaffinity设置线程的CPU亲和性
#define _GNU_SOURCE #include <sched.h> #include <stdlib.h> #include <sys/syscall.h> // syscall int main(void) { pid_t tid; int i, nrcpus; cpu_set_t mask; unsigned long bitmask = 0; CPU_ZERO(&mask); CPU_SET(0, &mask); /* add CPU0 to cpu set */ CPU_SET(2, &mask); /* add CPU2 to cpu set */ // get tid(线程的PID,线程是轻量级进程,所以其本质是一个进程) tid = syscall(__NR_gettid); // or syscall(SYS_gettid); /* Set the CPU affinity for a pid */ if (sched_setaffinity(tid, sizeof(cpu_set_t), &mask) == -1) { perror("sched_setaffinity"); exit(EXIT_FAILURE); } exit(EXIT_SUCCESS); }
---------------------------------------------------------------------------------------------------------------------
参考文献:
http://www.yboren.com/posts/44412.html?utm_source=tuicool
http://www.ibm.com/developerworks/cn/linux/l-affinity.html
http://saplingidea.iteye.com/blog/633616
http://blog.csdn.net/ttyttytty12/article/details/11726569
https://en.wikipedia.org/wiki/Processor_affinity
http://blog.chinaunix.net/uid-23622436-id-3311579.html
http://www.cnblogs.com/emanlee/p/3587571.html
http://blog.chinaunix.net/uid-26651253-id-3342161.html
http://blog.csdn.net/delphiwcdj/article/details/8476547
http://www.man7.org/linux/man-pages/man3/pthread_setaffinity_np.3.html
http://www.man7.org/linux/man-pages/man3/pthread_attr_setaffinity_np.3.html
man CPU_SET taskset