[ofiwg] completion flags as actually defined by OFI
Dave Goodell (dgoodell)
dgoodell at cisco.com
Tue Apr 14 14:06:45 PDT 2015
On Apr 14, 2015, at 3:18 PM, Hefty, Sean <sean.hefty at intel.com> wrote:
>>> I believe the providers support this guarantee.
>> Including the clean shutdown case?
> For FI_TRANSMIT_COMPLETE:
> - The sockets and psm providers do not generate a completion until the remote side has processed the request and acknowledged the data.
> - Cisco needs to confirm the usnic provider behavior, but it's UD anyway. I believe it adheres to the description given for completions on unreliable endpoints.
For DGRAM the usnic provider will only generate completions when the data has been put on the wire by the NIC, corresponding to FI_TRANSMIT_COMPLETE. For MSG+RDM I think we are currently offering FI_COMMIT_COMPLETE, though there are some known issues in those code paths right now. Since those two EP types are currently software based, there's no reason we couldn't choose to offer FI_COMMIT_COMPLETE if we find that we are not currently providing that semantic.
More information about the ofiwg