[openib-general] basic IB doubt

[Caitlin Bestler]

Woodruff, Robert J wrote:
> Catlin wrote,
>> Another point, even if a vendor were to implement the firmware you
>> suggest, how does the Data Source know that it is safe to use just
>> RDMA Writes? The enabling firmware is in the Data Sink.
> 
> Huh, don't understand the question.
> 
>> Applications certainly do not want to have to validate the model of
>> the RNIC that they are connected with.
> 
> If ISVs want to use an RNIC that does not support this
> technique, then they will have to implement their completion
> checking another way, which will be slower, so the hardware
> NICs that do not support this fast polling completion
> technique will be at a competitive disadvantage. Sometimes
> you can lead a horse to water, but you can't make then drink.

The benefit of "last byte RDMA Write ordering" would be to sipmlify
the logic of the remote peer doing the RDMA Writes. It does not benefit
the application doing the receiving.

The decision on whether or not to take the action that requires
a clean completion at the data sink must be taken by the data
source -- which has no method of knowing what vendor specific
features the remote peer has.

The whole point of using a standard protocol is to at least
define the optional features in a vendor independent way.