On 11/7/07, <b class="gmail_sendername">Richard Frank</b> <<a href="mailto:richard.frank@oracle.com">richard.frank@oracle.com</a>> wrote:<div><span class="gmail_quote"></span><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
<br>I'm pretty sure the code is not doing this - we should update the<br>comments - and get our the design docs as you keep reminding us to do.</blockquote><div><br>
I don't understand the code to the extent of saying whether or not it is done.<br>
</div><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">One thing we might want to consider is - if the IB transport knows that<br>we have a single HCA configured - then we can do away with the current
<br>HA ACKing - as it is only needed for HA with multiple adapters ! This<br>sounds a bit off - but consider all the single HCA boards out<br>there....running the HA acking is pure overhead.</blockquote><div><br>
I would not take this path - of having different code running for the
case you have one hca with one port, one hca with two ports, two hcas,
etc. Yet again I think you raise here an arguments against acking and
in favour of flow control, since the latter covers the first and is
needs in all schemes.<br>
</div><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">>> A couple of additional optimizations for our existing flow control<br>>> would be to 1) add srq support - this will reduce the possibility of
<br>>> RNRs 2) use an rdma write - vs - message for the ack to remove the<br>>> requirement of having a recv buffer posted to handle the ack.<br><br>> Can you elaborate a little on RDMA ACKs vs SEND ACKs? I guess you
<br>> don't mean to rdma-write-with-immediate since this also consumes a WR<br>> at the receiver side. If it just rdma-write, how would the sender be<br>> notified on the ACK reception, would it do polling-on-memory?
<br></blockquote><div><br>
</div><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">RDMA back the ack frame for an RC. And yes, on the send side - when a<br>send completes, and or when the CQ is drained, or when a send is
<br>initiated - check the current ACK hwm for the RC and release send<br>buffers which have been acked.</blockquote><div><br>
OK, thanks for the clarification. What is the status of 1 & 2 which you mention here, are they under implementation? <br>
<br>
Or.<br>
</div></div><br>