Sliding Window Protocol at the Transport Layer

The sliding window protocol combines pipelined transmission with reliability and flow control. Instead of waiting for an ACK after every segment (stop-and-wait), both sender and receiver maintain a window — a contiguous range of sequence numbers that may be outstanding at any instant.

Sender side

• The send window covers [SND.UNA … SND.NXT + usable space]: every sequence number inside it may be transmitted immediately, without waiting for an acknowledgement, so the link stays busy (pipelining).
• The window slides right only when a cumulative ACK arrives, which frees sequence space for new data.
• The usable window is min(cwnd, rwnd) — the congestion window (network congestion control) and the receiver's advertised window (flow control) — so the sender can overflow neither the network nor the receiver's buffer.
• On a time-out or 3 duplicate ACKs, the unacknowledged part of the window is retransmitted (Go-Back-N retransmits the whole window; Selective Repeat / TCP retransmits only the missing segments, optionally with SACK).

Receiver side

• The receive window is the set of sequence numbers it is currently willing to accept, decided by its free buffer space.
• This size is advertised to the sender in every segment (TCP Window field, 16 bits) — this is TCP's flow-control mechanism.
• The receiver returns cumulative ACKs; out-of-order segments are buffered (TCP, Selective Repeat) or discarded (Go-Back-N).

Example — window size = 4

Sender:   [ 1  2  3  4 ]                 initial window: 4 unacked segments
            send 1,2,3,4  ---------------------------->   Receiver accepts 1-4
ACK for 1 arrives ->  [ 2  3  4  5 ]      window slides, slot freed
ACK for 1,2,3 arrives ->       [ 4  5  6  7 ]
(ACK lost / time-out)  ->  retransmit from SND.UNA onward

Comparison of window-based protocols

Advantages: high link utilisation, natural place for flow control and error recovery, and it bounds the number of unacknowledged bytes. Limitation: the window must never exceed the sequence-number space (hence the 2ⁿ⁻¹ / 2ⁿ − 1 limits), or duplicate sequence numbers would be ambiguous.