>> For Heroic latencies on IB we would need to use small message RDMA and >> poll each peers dedicated memory region for completion. > > Well, I tried to play around with the eager_limit, min_rdma, etc. I did > not see the latency of messages of a given size be lowered by changing > the tresholds to make them rdma'd. Rather the opposite (which was my > initial expectation, actually). May be I just misunderstood the whole > set of tunables. My understanding was that messages under the eager > limit would never be rdma'd by definition, and that the others would or > would not be, depending on the min_rdma_size. >
The current code always uses send/recv semantics for short/eager messages. As Galen indicated, we will be adding support for short message rdma write, but this is generally not scalable, and would only be viable for a small working set of peers. Regards, Tim