Pipelining, Sliding Windows, and Bandwidth-Delay Product

Pipelining

To improve sender utilization, a protocol can use pipelining rather than stop-and-wait, permitting multiple packets to be sent before receiving ACKs.

Pipelining needs a larger sequence-number range because every in-flight packet needs a unique sequence number. Senders and receivers also need more storage for buffering packets.

Two basic approaches provide error recovery for pipelining:

1. Go-Back-N
2. Selective Repeat

Sliding Windows

The send window limits the number of unacknowledged packets in flight. $base$ identifies the oldest unacknowledged packet, while $nextseqnum$ identifies the next packet to send. With window size $N$, new data can be sent only if $nextseqnum < base+N$. Acknowledgments move the window forward.

Bandwidth-Delay Product and Utilization

BDP is the data a link can transmit within one propagation RTT. Here RTT excludes the first packet's serialization; ACK serialization and processing are negligible, packets have equal length, and no loss occurs:

To avoid a transmission gap before the first ACK returns, $N(L/R)\ge RTT+L/R$. The smallest integer window is:

Real networks also face loss, receiver-buffer limits, and congestion. A larger window cannot increase throughput indefinitely.