我应该何时终止 MPI 子进程？

Question

我正在使用 MPI 在 C++ 中实现一个算法。有许多文件需要处理。这是我的设计：

int main()
{
  MPI_Init();
  MPI_Comm_rank(MPI_COMM_WORLD, &my_rank);
  MPI_Comm_rank(MPI_COMM_WORLD, &nproc);

  MPI_Barrier(..);
  if(my_rank == 0)
  {
    for (each file to be processed)
      {
         Read in file content;
         MPI_Send data to child processes;
         process partial data on root process;
         MPI_Recv data processed by child processes;
         combine processed data from root and children;
      }
   }
   else
   {
      MPI_Recv data from root;
      process received data;
      MPI_Send processed data to root;
      MPI_Finalize();
    }

//only root process reaches here
MPI_Finalize();

}

当只有一个文件要处理时，程序运行完美。但是，如果我有超过 1 个文件要处理，它将停留在第二个文件中。而且似乎没有子进程可用于从根接收新数据。我认为这是因为我在处理第一个文件后终止了子进程。但是如果我在 else 块中注释掉 MPI_Finalize() ，程序将在处理第一个数据文件后退出，并出现错误：

mpirun has exited due to process rank 1 with PID 2003 on
node c301-115 exiting improperly. There are three reasons this could occur:

1. this process did not call "init" before exiting, but others in
the job did. This can cause a job to hang indefinitely while it waits
for all processes to call "init". By rule, if one process calls "init",
then ALL processes must call "init" prior to termination.

2. this process called "init", but exited without calling "finalize".
By rule, all processes that call "init" MUST call "finalize" prior to
exiting or it will be considered an "abnormal termination"

3. this process called "MPI_Abort" or "orte_abort" and the mca parameter
orte_create_session_dirs is set to false. In this case, the run-time cannot
detect that the abort call was an abnormal termination. Hence, the only
error message you will receive is this one. 

在这种情况下，有没有办法为子进程重置 MPI 实例？在哪里完成子进程的最佳位置？

Answer 1

您需要第二个 for 循环让工作人员等待新的分配，这样他们就不会立即终止。

这样想：你有 N 个人同时工作。你不知何故决定，其中一个名字为“0”的人有一项特殊的工作，即为所有其他人分配工作。你给出了每个人都应该做什么的准确规范。你写的代码在你的代码中看起来是这样的，这意味着

for(file in files)
   send job

对于名字为“0”的人

process one job

对于所有其他人。你希望其他人做的是：

for(file in files)
   process job

这应该反映在您的代码中，可能类似于：

int main()
{
  MPI_Init();
  MPI_Comm_rank(MPI_COMM_WORLD, &my_rank);
  MPI_Comm_rank(MPI_COMM_WORLD, &nproc);

  MPI_Barrier(..);
  if(my_rank == 0)
  {
    for (each file to be processed)
      {
         Read in file content;
         MPI_Send data to child processes;
         process partial data on root process;
         MPI_Recv data processed by child processes;
         combine processed data from root and children;
      }
   }
   else
   {
      for (each file to be processed)
        {
          MPI_Recv data from root;
          process received data;
          MPI_Send processed data to root;
          MPI_Finalize();
        }
    }

  //only root process reaches here
  MPI_Finalize();
}

旁注：您提议的极端等级结构有什么好处？如果每个工作线程都在自己的文件上工作不是更好吗？

Answer 2

请记住，您的流程需要保持“同步”。也就是说，对于每个发送都应该有一个接收等等，因此您需要在子进程中拥有一个循环，就像在根进程中一样。

这样做的一种方法是在程序开始时，让根进程将要处理的文件数发送给所有子进程。然后让它们像根进程一样循环。