TCPNet is supposed to use buffering which allows for faster data transmission rates. I see however that in some example programs/documentation, the handler of the TCP_STATE_CONNECT state in the send function checks if the previous frame (<1460 bytes) was acknowledged by the peer (by "tcp_check_send"). The question is: if my heap is large enough, can I send out large amounts of data that would later be acknowledged or do I need to wait for the acknowledgement of the large send frame?
It's in the manual:
www.keil.com/.../rlarm_tn_using_tcpsoc_sendtcp.htm
Well, yes, the design is inefficient. As the manual says, this is done for simplicity. If this is important to you, consider other TCP/IP stacks. For example, lwip does not have this limitation.