'How to use non blocking MPI-IO to overlap conputation and file operation?

I use MPI_File_iwrite with daos file system(a distribute file system).

    MPI_FIle_iwrite();
    //computation, time is long enough to cover the write operation)
    MPI_wait();  //to check if the MPI_FIle_iwrite is complete

Here the distribute file system (one server node and three storage nodes) can act as a coprocessor, so there is no need to use multi-threads. Since MPI_FIle_iwrite is a nonblocking operation, the IO and the computation should overlap, that is, the MPI_wait() should immediately return. Howerever, MPI_wait() cost a long time (just the same time the write operation should take). Why?

When I use MPI_Test, the MPI_Test also take a long time. Actually it should return immediately no matter if the operation is complete or not.

This problem is similar to mine

nonblockingfilewritermpi-io


Sources

This article follows the attribution requirements of Stack Overflow and is licensed under CC BY-SA 3.0.

Source: Stack Overflow

Solution Source

Related Questions