[Mpi-forum] MPI IO

[Amarpal Singh Kapoor]

Hello All,

I intend to solve a Finite Element problem using MPI and PETSc. I have a
conceptual problem with reading a mesh file in parallel with MPI IO.

My mesh file structure is as follows (3 blocks):

!Block1: Connectivity data

1 10 11 12 13   !ElementNumber, Node1, Node2, Node3, Node4   <element 1 is
connected to nodes 10, 11, 12 and 13>

2 14 15 16 17

.

.

.

****   ****   ****   ****

!Block2: Coordinate data

1 0.0 0.0 0.0  !Node1, x, y and z

2 0.0 0.0 1.0

.

.

.

****   **.**   **.**   **.**

!Block3: Boundary data

*   *   *  *  *

*   *   *  *  *

*   *   *  *  *  .

.

.

.

*   *   *  *  *

The goal is to read this single file parallely in multiple processors.
Parallelization type = MPI. No multithreading involved. I'll be using a
Lustre parallel file system.

I understand the idea summarized in the figure below:

Unlike in the figure above I have a file which is different in structure
(i.e. formatted with blocks of different size and datatypes) and I don't
get how different blocks of data can be read in parallel. Should there be a
synchronization call like MPI_BARRIER, after every block is completed on
all processors? Also, if there is a fortran example, I'll be really happy
to have it. Any suggestions are more than welcome.

Many Thanks,

Amar
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mpi-forum.org/pipermail/mpi-forum/attachments/20130615/724c8e64/attachment.html>