bosilca

What is the `first domain` ? One could argue that it should belong to the location where it has the most resources bound to ? Or where are less resources...

The simple fix, aka strict matching, is relatively easy to do by changing `hwloc_bitmap_intersects` to `hwloc_bitmap_isequal` in `opal_hwloc_compute_relative_locality` (opal/mca/hwloc/base/hwloc_base_util.c:638). The more complex matching that @edgargabriel proposed will require a complete...

Strictly the same binding at a specific level. I gave the outcome for the example provided here few comments above. On Genoa with the bindings provided by the user (`-np...

@mshanthagit as @edgargabriel noticed my example was incorrect. It should now be fixed: Rank 2 and 5 will be alone in their own communicator.

Honestly, I think that users binding processes as in the example here (overlapping several domains), deserve what they get and any split type is good, for as long as it...

There is only so much we can do automatically in the MPI library. For everything else, the users can fall back to either a manual `MPI_Comm_split` or to a guided...

> @bosilca does that mean that a process that does not fulfill this criteria (i.e. utilize that specific hardware resource type instance, and no other instance of the same hardware...

`UCX_NET_DEVICES` is your friend here. Set it to the interface you do intent to use, and this issue shall go away.

I think I see the problem: [`uct_tcp_query_devices`](https://github.com/openucx/ucx/blob/4fcffefedc51c62ab7ad5e1628da31c8f007aba1/src/uct/tcp/tcp_iface.c#L924) scans through all the interfaces, build a list of active and non-bridged interfaces and then trim it to the user requested devices. On...

It should not matter until the end of the iterator. We don't have to police every dep type, only that the end we have one.