C ++在g ++和msvc之间的文件读写时间不同

Question

I have a txt file with a few million rows (~80MB). I have to write a program in Visual Studio 2015, which reads the content of the file, sorts the rows, then write the result into another file. On my notebook there are two operating systems(Windows 7 and Ubuntu 15.04). First I wrote it on Ubuntu using g++, then I compiled the same source in Visual Studio 2015. I'm measuring these three operations times.

The results are:

Ubuntu(Ext3 partition)

• read: ~1s
• sort: ~3.2-3.4s
• write: ~1s

Ubuntu(run on NTFS partition)

• read: ~1s
• sort: ~3.2-3.4s
• write: ~4.7s

Windows(NTFS partition)

• read: ~5.5-6.0s(without optimisation it took over 2 minutes)
• sort: ~2.6s
• write: ~2.6-2.8s

Ubuntu: g++ -std=c++14 main.cpp(also tried with -O3 but the results were the same.

Windows: msvc compiler with -O3 optimisation

Tests were run on Asus K50AB.

So my question is that is it possible to get closer to the read/write times reached on Ubuntu, or msvc simply can't compile as efficient code as g++? Also I thought that the differences could be caused by different filesystems, but the read from Ubuntu on NTFS was the same speed.

auto t1 = std::chrono::high_resolution_clock::now();

std::ifstream is{ "rec.txt" };
std::ofstream os{ "res.txt" };

// number is the number of lines
std::vector<std::string> lines(number);

for (int i = 0; i < number; ++i)

    is >> lines[i];

auto t2 = std::chrono::high_resolution_clock::now();

std::cout << "read time: " <<
    std::chrono::duration_cast<std::chrono::milliseconds>(t2 - t1).count() << std::endl;


// sort elements


t1 = std::chrono::high_resolution_clock::now();

for (int i = 0; i < number; ++i)

    os << s[i] << '\n';

t2 = std::chrono::high_resolution_clock::now();

std::cout << "write time: " <<
    std::chrono::duration_cast<std::chrono::milliseconds>(t2 - t1).count() << std::endl;

Answer 1

There are a lot of things coming into play when doing file io, such as os caching, anti virus, Windows Search (yes). I would not attribute any differences in speed to Visual Studio vs gcc compilers just yet.

If you want to get higher speeds, make sure to use reasonably large file reads (I have no idea what the actual file io sizes will be in your case).

If you want to compare the speed of your program, don't mix measurements between sorting and file io. File io is slooooooooow.

You can see from your own measurements that the sorting itself takes approximately the same amount of time in your example or is actually faster using VS.

Read large chunks at the time from disk or possibly map it into memory.