Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

v5.0.x: docs: add some info about HPE CXI ofi provider #12163

Merged
merged 2 commits into from
Dec 21, 2023
Merged

v5.0.x: docs: add some info about HPE CXI ofi provider #12163

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
2e62456
docs: add some info about HPE CXI ofi provider
hppritcha Dec 13, 2023
212d8ae
docs: refine wording related to HPE CXI provider
hppritcha Dec 15, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 9 additions & 2 deletions docs/tuning-apps/networking/ofi.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,6 +18,7 @@ including (but not limited to):
* Cisco usNIC
* Cray uGNI
* Cornelius Networks Omni-Path
* HPE Slingshot 11

In general, the OFI-based components in Open MPI will auto-select
themselves as appropriate at run time.
Expand All @@ -43,8 +44,9 @@ communications:
``FI_EP_RDM`` endpoints, ``FI_TAGGED`` capabilities, and
``FI_ORDER_SAS`` ordering).

#. ``ofi`` BTL: Available since Open MPI v4.0.0, this component is
used for one-sided MPI communications (e.g., ``MPI_PUT``). The
#. ``ofi`` BTL: Available since Open MPI v4.0.0, this component is primarily
intended for one-sided MPI communications (e.g., ``MPI_PUT``). It
can also support BTL send/recv operations.
``ofi`` BTL requires that the Libfabric provider support reliable
datagrams, RMA and atomic operations, and remote atomic completion
notifications (specifically: ``FI_EP_RDM`` endpoints, ``FI_RMA``
Expand Down Expand Up @@ -76,6 +78,11 @@ list. For example:

shell$ ompi_info --param mtl ofi --level 9

.. important:: When using the HPE CXI provider and ``mpirun`` as the job launcher,
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Just to be clear: this isn't entirely accurate. You could also simply add --map-by :nouselocal to the mpirun cmd line.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

yeah, if you want to get charged for N+1 vs N nodes where N is the number of nodes available for your MPI processes.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I understand your point - however, your text doesn't indicate that you have a choice. You state that you must use the specified option - and that simply isn't true.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That's a fair point. It's probably worth pointing out the subtlety here -- i.e., what does this parameter mean and do, vs. what the user's options are if they don't use this MCA param.

the PRTE ``ras_base_launch_orted_on_hn`` MCA parameter must be set to 1.
This can be done by adding ``--prtemca ras_base_launch_orted_on_hn 1`` to the job launch
command line.

For more information refer to the `Libfabric web site
<https://libfabric.org/>`_.

Expand Down