[libfabric-users] Scalable endopint issue
sean.hefty at intel.com
Fri Dec 10 08:50:58 PST 2021
> Question : What is the expected use case of shared receive contexts? If I can't use
> them one per thread (at least for tagged messages), then when would I use them. If I
> had only unexpected messages, then one per thread might work, but you'd need to be very
> carefully to ensure all threads were polling right up to the end, otherwise some rx's
> might end up on cq's that were idle ...
For non-tagged messages, it allows traffic to be distributed to multiple threads, like you're wanting to do. Someday we can add flow steering to the API, but given the lack of HW that supports scalable contexts, this just hasn't been a priority.
So, the only 'flow steering' we currently have are named rx contexts. That's send side steering, however.
Howard might have more details on the gni implementation.
More information about the Libfabric-users