[libfabric-users] FI_EP_MSG on cray

[Howard Pritchard]

John,

For the GNI provider test run failures, what kind of allocation are you
asking for on the machine prior to
running run_gnitest?  Do you have exclusive access to the node you are
using?

It might be helpful to
export UGNI_DEBUG=10
or higher and then
run the script again.  Please post the output to the list.

Are you using the theta system for this work?

Howard



2017-02-13 11:04 GMT-07:00 Hefty, Sean <sean.hefty at intel.com>:

> > libfabric:ofi-rxm:core:ofi_check_ep_attr():397<info> Unsupported
> > endpoint type
> >
> > libfabric:ofi-rxm:core:ofi_check_ep_attr():398<info> Supported:
> > FI_EP_RDM
> >
> > libfabric:ofi-rxm:core:ofi_check_ep_attr():398<info> Requested:
> > FI_EP_MSG
> >
> >
> >
> > which I presume means that EP_MSG type is not supported.
> >
> >
> >
> > Is this likely to be available any time soon? I see it as a ‘planned’
> > feature, but I wonder what the time scale is.
>
>
> Ofi-rxm is a 'utility' provider that implements RDM EPs over MSG EPs.  So
> that behavior is expected for the RXM code.  Note that before the next
> release, we will remove ofi-rxm from being reported as an actual provider.
>
> You can use the FI_PROVIDER environment variable to limit which provider
> will be used.
>
> - Sean
> _______________________________________________
> Libfabric-users mailing list
> Libfabric-users at lists.openfabrics.org
> http://lists.openfabrics.org/mailman/listinfo/libfabric-users
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/libfabric-users/attachments/20170213/d4e72ad4/attachment.html>