要读取一个文件，哪个会更快？使用 fstream、FILE* 还是内存映射？

Question

这里是我的项目配置：vs2013,win32,Debug。

我想知道对于不同的文件大小，3种文件读取方法中的哪一种faster.They是c++风格的fstream，c风格的文件读写和内存映射。

但是执行后，这是我的结果：

File Size 1225284
fstream time 47
c file pointer time 0
memory mapping time 0

File Size 14856192
fstream time 15
c file pointer time 0
memory mapping time 47

File Size 97198080
fstream time 16
c file pointer time 0
memory mapping time 265

File Size 1259530844
fstream time 31
c file pointer time 16
memory mapping time 11138

似乎对于stream和FILE*的读取，读取一个文件所需的时间不会随着文件大小increases.But的内存映射而增加，这种true.This现象很奇怪。
因为在我看来，对于大文件，内存映射会更快。这是我的代码：

string ifile = "M:/Thesis/FileReadCmp/1.txt";
string os = "M:/Thesis/FileReadCmp/new_cmp1.txt";
int page_size = 2 * 1024 * 64 * 1024;//128M
for (int j = 0; j < 100; ++j){
    os[os.size() - 5] = '1' + j;
    ofstream o(os);
    for (int i = 0; i < 4; ++i){
        ifile[ifile.size() - 5] = '1' + i;
        ifstream in(ifile);

        in.seekg(0, ios::end);
        o << "File Size " << in.tellg() << endl;
        o << endl;
        in.close();

        //using fstream to read file
        long long st = GetTickCount();
        in.open(ifile);
        char c;
        while (in >> c){
            ;
        }
        in.close();             
        long long et = GetTickCount();
        o << "fstream time " << et - st << endl;

        //using FILE* to read file
        st = GetTickCount();
        FILE* cpf = fopen(ifile.c_str(), "r");
        char cc = fgetc(cpf);
        while (cc != EOF)
        {
            cc = fgetc(cpf);
        }
        fclose(cpf);
        et = GetTickCount();
        o << "c file pointer time " << et - st << endl;

        //using memory mapping to read file
        const char* pc = ifile.c_str();
        st = GetTickCount();
        HANDLE hFile = CreateFile(pc, GENERIC_WRITE | GENERIC_READ, 0,
            NULL, OPEN_EXISTING, FILE_ATTRIBUTE_NORMAL, NULL);
        int file_size = GetFileSize(hFile, NULL);
        HANDLE hFileMap = OpenFileMapping(FILE_MAP_READ | FILE_MAP_WRITE, FALSE,
            TEXT("SharedData"));

        if (hFileMap == NULL){
            // if no such object,create a file mapping object
            hFileMap = CreateFileMapping(hFile, NULL, PAGE_READWRITE,
                0, 0, TEXT("SharedData"));
        }

            int rem_file_size = file_size;
            int offset = 0;
            while (rem_file_size > page_size){
                PVOID pvFileView = MapViewOfFile(hFileMap, FILE_MAP_WRITE, 0, offset, page_size);
                char* asc_dex = (char*)pvFileView;
                for (int i = 0; i < page_size; ++i){//, c = 0++c
                    char c = asc_dex[i];
                }
                //UnmapViewOfFile(pvFileView);
                offset += page_size;
                rem_file_size -= page_size;
            }
            PVOID pvFileView = MapViewOfFile(hFileMap, FILE_MAP_WRITE, 0, offset, rem_file_size);
            char* asc_dex = (char*)pvFileView;
            for (int i = 0; i < rem_file_size; ++i){//, c = 0++c
                char c = asc_dex[i];
            }
            UnmapViewOfFile(pvFileView);

        CloseHandle(hFileMap);
        CloseHandle(hFile);

        et = GetTickCount();
        o << "memory mapping time " << et - st << endl;
        o << endl;
    }
}

Answer 1

最有效的方法是保持输入流流畅。通常，这意味着读取大块数据（每个请求的数据更多）。

一种方法是使用多个缓冲区（在互联网上搜索 "double buffering"）和多个线程。

读取线程将读取并填满一个缓冲区，然后发出数据准备就绪的信号。读取线程继续填充下一个缓冲区。

处理线程（等待读取信号）唤醒并开始处理缓冲区，然后输出数据。

一个objective是使用尽可能多的大缓冲区来调整读取数据的速度与处理速度。同样，这个想法是数据读取在处理方面是连续的。

Answer 2

可能是打开的问题mode.The内存映射的默认打开方式是二进制。

当我把in.open(ifile);改成in.open(ifile, ios::binary);；FILE* cpf = fopen(ifile.c_str(), "r");改成FILE* cpf = fopen(ifile.c_str(), "rb");3种方法真正的运行次变成按照theory.They如下：

File Size 1225284
fstream time 297
c file pointer time 93
memory mapping time 0

File Size 14856192
fstream time 3167
c file pointer time 1045
memory mapping time 47

File Size 97198080
fstream time 20779
c file pointer time 6833
memory mapping time 281

File Size 1259530844
fstream time 268696
c file pointer time 88406
memory mapping time 11216

要读取一个文件，哪个会更快？使用 fstream、FILE* 还是内存映射？

To read a file,which one will be faster?using fstream,FILE* or memory mapping?

c++

file

stream

memory-mapping