Tang, Changqing wrote:
Or:
I think I find the SPEC part talking about zero-length message.
I just search zero-length.
For rdma write, the receiver side does not need to do anything,
for send, a receive
WR is need in receiver side. And that makes sense.
Indeed. However, i am not sure that for the scales of K ranks job, i
would run to implements zero-length-rdma-write keep alive protocol
putting O(K^2) messages on the wire before making sure its really needed
to fulfill the functional demand.
Or.
_______________________________________________
general mailing list
[email protected]
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
