We should define a set of performance criteria / tests to ensure we do
not impact our current performance with IB...
An alternative would be to push this into an IWARP specific
module....and if works well there - we might then want to move it to
generic RDS layer ? As an example, the TCP transport for RDS - handles
flow control internally..
Jon Mason wrote:
As part of my effort to get RDS working for iWARP, I will be working on the
RDS flow control. Flow control is needed for iWARP due to the fact that
iWARP connections terminate if there is no posted recv for an incoming
packet. IB connections do not have this limitation if setup in a certain
way. In its current implementation, RDS sets the connection attribute
rnr_retry to 7. This causes IB to retransmit until there is a posted recv
buffer.
Using a credit based flow control mechanism, we can ensure there will be a
posted recv for every incoming packet (thus laying part of the foundation of
allowing iWARP to work). Also, it will reduce unnecessary IB transport
traffic (at the expense of maintaining the credit schema).
I am still in the very early stages of implementing this. So any pointers to
RDS documentation (or a RDS git tree) would be very helpful. I have a small
IB setup to test this on, so anyone willing to test it when I am done would
be helpful as well.
Thanks,
Jon
_______________________________________________
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
_______________________________________________
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
