linux磁盘IO测试

jinkun00

2013-08-19

man 2 fsync知fsync会真正把文件的数据写回磁盘，

但是在3.2的kernal上测试发现fsync最好集中一次做掉，

可能fsync执行过程不会判断每个块是否正在修改，都去刷盘。（有待确认）

const int SIZE = 1 << 24;                                                       
char buf[SIZE];                                                                 
                                                                                
void test(int size) {                                                           
    int fd = open("txt", O_WRONLY | O_CREAT | O_TRUNC, S_IRWXU);                
    int count = SIZE / size;                                                    
    for (int i = 0; i < count; ++i) {                                           
        write(fd, buf, size);                                                   
        fsync(fd);                                                              
    }                                                                           
    //fsync(fd);                                                                
    close(fd);                                                                  
}

[size] time consume:xxxms

此时测得

[131072] time consume: 5531ms

[262144] time consume: 2718ms

[524288] time consume: 1464ms

[1048576] time consume: 926ms

[2097152] time consume: 628ms

[4194304] time consume: 404ms

[8388608] time consume: 305ms

[16777216] time consume: 256ms

16777216 = 2^24

但是如果用下面的fsync发现

[131072] time consume: 331ms

[262144] time consume: 264ms

[524288] time consume: 239ms

[1048576] time consume: 264ms

[2097152] time consume: 239ms

[4194304] time consume: 264ms

[8388608] time consume: 239ms

[16777216] time consume: 263ms

还是mmap给力

....

[64] time consume: 41ms

[128] time consume: 42ms

[256] time consume: 41ms

[512] time consume: 41ms

[1024] time consume: 41ms

[2048] time consume: 41ms

[4096] time consume: 33ms

[8192] time consume: 33ms

[16384] time consume: 33ms

[32768] time consume: 33ms

[65536] time consume: 41ms

[131072] time consume: 42ms

[262144] time consume: 41ms

[524288] time consume: 41ms

[1048576] time consume: 50ms

[2097152] time consume: 74ms

[4194304] time consume: 107ms

[8388608] time consume: 141ms

struct node {                                                                   
    int fd_, size_;                                                             
    node () { }                                                                 
    node (int fd, int size) {                                                   
        fd_ = fd;                                                               
        size_ = size;                                                           
    }                                                                           
};                                                                              
void test_prefetch(node t) {                                                    
    char *p = (char*)mmap(NULL, t.size_, PROT_WRITE, MAP_SHARED, t.fd_, 0);     
    if (p == MAP_FAILED) {                                                      
        perror("mmap");                                                         
        exit(1);                                                                
    }                                                                           
    memcpy(p, buf, t.size_);                                                    
    msync(p, t.size_, MS_SYNC);                                                 
    munmap(p, t.size_);                                                         
}

前面很多size虽然不一样，但是写入磁盘的时间却是一样的，说明磁盘的块特性，

fdisk -l

Disk /dev/sda: 500.1 GB, 500107862016 bytes

255 heads, 63 sectors/track, 60801 cylinders, total 976773168 sectors

Units = sectors of 1 * 512 = 512 bytes

Sector size (logical/physical): 512 bytes / 4096 bytes

I/O size (minimum/optimal): 4096 bytes / 4096 bytes

Disk identifier: 0xe0940810

linux系统