Connections. for Example, with Our Network with Three a New Architecture for Packet Switch- Ing Network Protocols, In

connections, with a buer size of 32 and a pipe size of 50 packets, the relative shares of throughput were 31%, 39%, and 30%. This anomalous behavior can be removed by simply changing the congestion avoidance increase algorithm to read: cwnd += 1 / wnd With this change, bcwndc always increases by one in every epoch. For those who are interested in experimenting with the above algorithm, the modied TCP code that we are currently exploring is described below. In real BSD TCP code, cwnd is in terms of bytes, instead of packets. The above modication requires a one-line change in tcp_input() procedure: replace u_int incr = tp->t_maxseg; if (tp->snd_cwnd > tp->snd_ssthresh) incr = MAX(incr * incr / tp->snd_cwnd, 1); by u_int incr = tp->t_maxseg; if (tp->snd_cwnd > tp->snd_ssthresh) incr = MAX(incr / (tp->snd_cwnd / incr), 1); References [1] R. Braden (editor). Requirements for Internet hosts-communication layers, RFC-1122, Octo-ber 1989.