[ofa-general] detecting "recv buffer not available" events

[Talpey, Thomas]

At 05:21 PM 3/4/2008, Ralph Campbell wrote:
>There is no generic counter you can query to find the number
>of RNR retires. The best you can do is set the IBV_QP_RNR_RETRY
>parameter to zero and see if you get any IBV_WC_RNR_RETRY_EXC_ERR
>error completions. If IBV_QP_RNR_RETRY is N, the HCA
>will do up to N=0..6 retries for you (or if it is 7, an infinite
>number of times).

The NFS/RDMA client sets this to 0 when invoking rdma_connect(),
yet no failures are seen on IB sends even when the server fails to
promptly post receives.

So, does the setting mean anything on the receiving end? I.e., no matter
what the NFS/RDMA server sets it to, shouldn't we expect the client's
sends to fail if they don't land on the first try in a receive buffer there?
That's why the client zeroes rnr_retry - the RPC/RDMA protocol manages
this so we don't want the transport "helping".

Tom.

>
>On Tue, 2008-03-04 at 15:25 -0600, Steve Wise wrote:
>> All,
>> 
>> This question is regarding SENDs arriving when no RECV buffer is posted. 
>>   For IB, somehow the transport can deal with this and force the sender 
>> transport to retransmit.
>> 
>> First, is my statement above correct?
>> 
>> If so, How can I detect that this is happening on an IB RC?  IE is there 
>> some stat I can query?
>> 
>> Thanks,
>> 
>> Steve.
>> _______________________________________________
>> general mailing list
>> general at lists.openfabrics.org
>> http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
>> 
>> To unsubscribe, please visit 
>http://openib.org/mailman/listinfo/openib-general
>
>_______________________________________________
>general mailing list
>general at lists.openfabrics.org
>http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
>
>To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general