Skip to content

Commit

Permalink
Merge pull request #12 from edgargabriel/pr/ompio-doc-minor-fixes
Browse files Browse the repository at this point in the history
docs/ompio: minor fixes
  • Loading branch information
jsquyres authored Dec 24, 2022
2 parents 0ce1551 + 0563250 commit 541bfc6
Showing 1 changed file with 4 additions and 10 deletions.
14 changes: 4 additions & 10 deletions docs/tuning-apps/mpi-io/ompio.rst
Original file line number Diff line number Diff line change
Expand Up @@ -100,8 +100,8 @@ important ``fcoll`` components are:
The ``sharedfp`` framework provides a different implementation of the
shared file pointer operations depending on file system features.

* ``lockfile``: this component will be used on file system which
support for file locking.
* ``lockedfile``: this component will be used on file systems which
support file locking.

* ``sm``: component used in scenarios in which all processes of the
communicator are on the same physical node.
Expand Down Expand Up @@ -131,18 +131,12 @@ operation are listed below:
Tuning this parameter has a very high impact on the performance of
collective operations.

.. note:: Be sure to also see recommendations for tuning collective
operations.

#. ``io_ompio_num_aggregators``: Number of aggregators used in
collective I/O operations. Setting this parameter to a value
larger zero disables the internal automatic aggregator selection
logic of OMPIO. Tuning this parameter has a very high impact on
the performance of collective operations.

.. note:: Be sure to also see recommendations for tuning collective
operations.

#. ``io_ompio_grouping_option``: Algorithm used to automatically
decide the number of aggregators used. Applications working with
regular 2-D or 3-D data decomposition can try changing this
Expand Down Expand Up @@ -190,8 +184,8 @@ Using GPU device buffers in MPI File I/O operations

OMPIO supports reading and writing directly to/from GPU buffers using
the MPI I/O interfaces. Using this feature simplifies managing buffers
that are exclusive used on GPU devices, and hence the necessity to
implement a staging through host memory for file I/O operations.
that are exclusively used on GPU devices, and hence there is no need to
implement staging through host memory for file I/O operations.

Internally, OMPIO splits a user buffer into chunks for performing the
read/write operation. The chunk-size used by OMPIO can have a
Expand Down

0 comments on commit 541bfc6

Please sign in to comment.