Olaf Kirch wrote:
In order to be able to test RDS (and RDS over RDMA) faithfully on standalone machines, we want loopback traffic to use the IB transport if possible.
Olaf,
Beyond the details of this patch, one thing which should be on the table here is the IB RC LID matching rules stating that when an RC QP is configured it is set with <SLID, DLID> and if a packet is received where its LRH does not match these lids, its dropped. In other words, if through the test, the standalone machine becomes attached to IB fabric and LIDs are set to the ports by the SM, the loopback connection would get broken. Generally speaking, with RDS this should not be big deal, since it will reconnect, but I just wanted to make sure we all aware to this.
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
