> > I believe that any solution here should mimic the TCP/IP stack as > closely as possible. So I would rule out the re-use of a single port > for all active connections. > > > > I think TCP matches on the full tuple <src port, src ip, dst port, > dst ip>. We should be safe to re-use port numbers as long as some > other portion of the tuple changes. Maybe that can be added as part of > the port reservation/checking? > > > > At first the thought was to reuse ports as long as the dest IP between > rdma_id is different but is this complication really necessary? > RDMA_CM mimics socket API but wire protocol is different and source > port has no role in transporting a packet from QP to QP. Do you see a > real risk in reusing a port unconditionally? The port number is carried in the packet and provided as the source address to the other side. By re-using a single port number, a server will see multiple connections all reporting the same source address (port + src IP). The proposal will break iWarp. ��.n��������+%������w��{.n�����{���fk��ܨ}���Ơz�j:+v�����w����ޙ��&�)ߡ�a����z�ޗ���ݢj��w�f