How to optimize C++ I/O operations to improve performance?

To improve C I/O performance, several approaches can be taken: Use buffered I/O to group data to reduce the number of disk accesses. Use the mmap() system call to map files directly into memory to avoid frequent disk accesses. Use parallel I/O to perform I/O operations simultaneously on multiple threads or processes to increase throughput.

How to optimize C I/O operations to improve performance

I/O operations are critical to the performance of your application. In C, there are several ways to optimize I/O operations to improve performance.

1. Using Buffered I/O

Buffered I/O involves grouping data into large chunks and then writing or reading them from disk. This reduces the number of disk accesses, thereby improving performance.

#include <iostream>
#include <fstream>
#include <vector>

int main() {
  std::vector<int> data(1000000);
  std::ofstream file("data.bin", std::ios::binary);
  // 缓冲 1 MB 的数据
  file.rdbuf()->pubsetbuf(nullptr, 1024 * 1024);

  // 写入数据
  file.write((char*)&data[0], data.size() * sizeof(int));
  file.close();

  return 0;
}

2. Using mmap()

The mmap() system call allows you to map files directly into memory. This avoids frequent disk accesses, thus improving performance.

#include <sys/mman.h>
#include <fcntl.h>

int main() {
  // 打开文件
  int fd = open("data.bin", O_RDWR);
  // 将文件映射到内存
  void* data = mmap(nullptr, 1000000 * sizeof(int), PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
  
  // 操作数据
  ...

  // 取消映射
  munmap(data, 1000000 * sizeof(int));
  close(fd);

  return 0;
}

3. Using Parallel I/O

Parallel I/O involves performing I/O operations on multiple threads or processes simultaneously. This can improve throughput and reduce overall execution time.

#include <thread>
#include <vector>

int main() {
  std::vector<std::thread> threads;
  for (int i = 0; i < 4; i++) {
    threads.emplace_back([] {
      // 执行 I/O 操作
    });
  }

  for (auto& thread : threads) {
    thread.join();
  }

  return 0;
}

Practical case

The following is a practical case using C to optimize I/O operations. This program reads and writes large amounts of data from a file:

#include <iostream>
#include <fstream>
#include <vector>
#include <chrono>

using namespace std;

int main() {
  // 数据量
  const int dataSize = 1000000;

  // 使用缓冲 I/O
  {
    vector<int> data(dataSize);
    ofstream file("data.bin", ios::binary);
    file.rdbuf()->pubsetbuf(nullptr, 1024 * 1024);

    // 记录时间
    auto start = chrono::high_resolution_clock::now();
    // 写入数据
    file.write((char*)&data[0], data.size() * sizeof(int));
    auto end = chrono::high_resolution_clock::now();

    // 计算执行时间
    auto duration = chrono::duration_cast<chrono::milliseconds>(end - start);
    cout << "Buffered I/O duration: " << duration.count() << " ms" << endl;
  }

  // 使用 mmap()
  {
    vector<int> data(dataSize);
    int fd = open("data.bin", O_RDWR);
    void* dataPtr = mmap(nullptr, dataSize * sizeof(int), PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);

    // 记录时间
    auto start = chrono::high_resolution_clock::now();
    // 写入数据
    memcpy(dataPtr, &data[0], data.size() * sizeof(int));
    auto end = chrono::high_resolution_clock::now();

    // 取消映射
    munmap(dataPtr, dataSize * sizeof(int));
    close(fd);

    // 计算执行时间
    auto duration = chrono::duration_cast<chrono::milliseconds>(end - start);
    cout << "mmap() duration: " << duration.count() << " ms" << endl;
  }

  return 0;
}

Run this program and you'll notice that using mmap() is many times faster than buffered I/O.

The above is the detailed content of How to optimize C++ I/O operations to improve performance?. For more information, please follow other related articles on the PHP Chinese website!