Abstract: Method and apparatus for returning packet acknowledgements from a packet receiving node to a packet sending node of a packet network. During a predefined initial startup phase of a packet connection between the sending and receiving nodes, an acknowledgement is returned to the sending node for every group of X packets received from the sending node. After the expiration of the initial startup phase, an acknowledgement is returned for every W-N packets received from the sending node. W represents the maximum number of packets that can be in transit on the connection at any given time, N is an integer less than W. If packet congestion occurs, the algorithm reverts to returning an acknowledgement for every group of X packets.