Selective Repeat keeps the same full pipeline as Go-Back-N but makes the receiver smart. Instead of discarding frames that arrive before the gap is filled, it buffers them and acknowledges each frame individually. When a frame is lost, the sender retransmits only that one frame — the early arrivals are already safe in the receiver's buffer and get delivered once the missing frame fills the hole.
How it differs from Go-Back-N
1Receiver has a window > 1 and a buffer: out-of-order frames within the window are kept, not thrown away.
2Each frame gets its own individual ACK, so the sender knows exactly which frames are still missing.
3Sender keeps a separate timer per outstanding frame; only the frame whose timer expires is resent — selective retransmission.
4When the lost frame finally arrives, the receiver delivers it together with the buffered frames, all in order.
Both windows must be equal — and ≤ 2^(m−1)
Unlike Go-Back-N (receiver window 1), Selective Repeat needs sender and receiver windows of the same size, and that size must be at most `2^(m−1)` for m-bit sequence numbers. If the window were larger, a retransmitted old frame could carry a sequence number the receiver mistakes for a new one, and it would accept a duplicate as fresh data.
Least waste, most bookkeeping
Selective Repeat retransmits the fewest frames of the three ARQ schemes, but it pays for it: buffering on both ends, per-frame timers, and per-frame ACKs. Go-Back-N is simpler; Selective Repeat is leaner on a lossy link.
OperationTimeSpace
On one lost frame · window ≤ 2^(m−1); individual ACKs, per-frame timersresend 1 framesender N · receiver N
Check yourself
With m-bit sequence numbers, what is the maximum window size for Selective Repeat?
