malloc与缺页的一些的时间测量

malloc与缺页的一些的时间测量

时间函数

struct timespec
{
 time_t tv_sec;
 long int tv_nsec;
};
/*
1.which_clock参数解释

CLOCK_REALTIME:系统实时时间,随系统实时时间改变而改变,即从UTC1970-1-1 0:0:0开始计时,中间时刻如果系统时间被用户该成其他,则对应的时间相应改变

CLOCK_MONOTONIC:从系统启动这一刻起开始计时,不受系统时间被用户改变的影响

CLOCK_PROCESS_CPUTIME_ID:本进程到当前代码系统CPU花费的时间

CLOCK_THREAD_CPUTIME_ID:本线程到当前代码系统CPU花费的时间

*/
long sys_clock_gettime (clockid_t which_clock, struct timespec *tp);

这个函数最终会是系统调用，会进入内核，所以估计这个函数的执行时间会比较久，所以先测试这个函数的执行时间代码如下

#include <stdio.h>
#include <stdlib.h>
#include <time.h>
int main()
{
    struct timespec timeStart = {0,0},timeEnd = {0,0};
    clock_gettime(CLOCK_PROCESS_CPUTIME_ID,&timeStart);
    for(int i=0;i<1000;++i)
    {
        clock_gettime(CLOCK_PROCESS_CPUTIME_ID,&timeEnd);
    }
    printf("timeStart = %lus,%luns\n",timeStart.tv_sec,timeStart.tv_nsec);
    printf("timeEnd = %lus,%luns\n",timeEnd.tv_sec,timeEnd.tv_nsec);
    printf("ever sys_gettime call spend %luns\n",(timeEnd.tv_nsec-timeStart.tv_nsec)/1001);
    return 0;

}

最终测试结果摆动较大，在100ns至500ns间摆动，考虑到误差，取均值300ns

开始测试malloc

缺页不频繁的情况下，主要的开销是malloc的算法开销。

第一次测试由于malloc后立马free，所以缺页率非常低

#include <stdio.h>
#include <stdlib.h>
#include <time.h>
int main(int argc,char** argv)
{
    if(argc!=2)
    {
        fprintf(stderr,"error args\n");
        exit(-1);
    }
    void *p;
    int sz = atoi(argv[1]);
    long int sum = 0;
    struct timespec timeStart = {0,0},timeEnd = {0,0};
    for(int i=0;i<1000;++i)
    {
        clock_gettime(CLOCK_PROCESS_CPUTIME_ID,&timeStart);
        p = malloc(sz);
        clock_gettime(CLOCK_PROCESS_CPUTIME_ID,&timeEnd);
        sum += timeEnd.tv_nsec - timeStart.tv_nsec;
        free(p);
    }
    printf("malloc %dByte every time for 1000loops,%luns per\n",sz,sum/1000);
    return 0;

}
/*测试结果
wd@ub-linux:malloc$ ./a.out 256
malloc 256Byte every time for 1000loops,649ns per
wd@ub-linux:malloc$ ./a.out 1024
malloc 1024Byte every time for 1000loops,742ns per
wd@ub-linux:malloc$ ./a.out 4096
malloc 4096Byte every time for 1000loops,222ns per
wd@ub-linux:malloc$ ./a.out 4096
malloc 4096Byte every time for 1000loops,763ns per
wd@ub-linux:malloc$ ./a.out 5000
malloc 5000Byte every time for 1000loops,643ns per
wd@ub-linux:malloc$ ./a.out 5000
malloc 5000Byte every time for 1000loops,716ns per
wd@ub-linux:malloc$ ./a.out 5000
malloc 5000Byte every time for 1000loops,742ns per
wd@ub-linux:malloc$ ./a.out 5000
malloc 5000Byte every time for 1000loops,801ns per
wd@ub-linux:malloc$ ./a.out 5000
malloc 5000Byte every time for 1000loops,327ns per
wd@ub-linux:malloc$ ./a.out 5000
malloc 5000Byte every time for 1000loops,319ns per
wd@ub-linux:malloc$ ./a.out 5000
malloc 5000Byte every time for 1000loops,711ns per
wd@ub-linux:malloc$ ./a.out 5000
malloc 5000Byte every time for 1000loops,816ns per
wd@ub-linux:malloc$ ./a.out 5000
malloc 5000Byte every time for 1000loops,282ns per
wd@ub-linux:malloc$ ./a.out 50000
malloc 50000Byte every time for 1000loops,596ns per
wd@ub-linux:malloc$ ./a.out 50000
malloc 50000Byte every time for 1000loops,661ns per
wd@ub-linux:malloc$ ./a.out 500000
malloc 500000Byte every time for 1000loops,500ns per
wd@ub-linux:malloc$ ./a.out 500000
malloc 500000Byte every time for 1000loops,751ns per
wd@ub-linux:malloc$ ./a.out 500000
malloc 500000Byte every time for 1000loops,178ns per
wd@ub-linux:malloc$ ./a.out 500000
malloc 500000Byte every time for 1000loops,222ns per
wd@ub-linux:malloc$ ./a.out 500000
malloc 500000Byte every time for 1000loops,662ns per
wd@ub-linux:malloc$ ./a.out 500000
malloc 500000Byte every time for 1000loops,749ns per
wd@ub-linux:malloc$ ./a.out 500000
malloc 500000Byte every time for 1000loops,546ns per
wd@ub-linux:malloc$ ./a.out 500000
malloc 500000Byte every time for 1000loops,600ns per
wd@ub-linux:malloc$ ./a.out 500000
malloc 500000Byte every time for 1000loops,546ns per

*/

可以看到这种情况下，每次的时间在200ns到800ns间摆动，由于还加了sys_gettime的开销，所以最终的一次malloc的开销大概在200ns左右。注意，这是在没有缺页的情况下！！！！！

而如果发生了缺页，则时间非常之就，比如将上面的代码的free(p)注释掉，然后执行

wd@ub-linux:malloc$ ./a.out 4096
malloc 4096Byte every time for 1000loops,3855ns per

会发现时间变得特别巨大，直接跳到了3800ns+。这个参数是4096，就是说每次申请4k的内存，就是一页内存而不还给os，这样几乎每次申请都会出现缺页。可以看到缺页处理时间几乎是malloc本身算法执行时间的10倍。