fastrecovery

Fast Recovery, sometimes referred to as fastrecovery, is a TCP congestion control mechanism designed to quickly recover from packet loss detected by triple duplicate acknowledgments. It is part of the fast retransmit and fast recovery suite and is commonly associated with TCP Reno. The mechanism was introduced to improve performance compared with earlier Tahoe behavior, and has influenced subsequent congestion control variants, including NewReno and other modern algorithms.

Upon receiving three duplicate ACKs, a sender using Fast Recovery reduces its slow-start threshold (ssthresh) to half of its current congestion window (cwnd) and sets cwnd to ssthresh plus three MSS values. This allows the sender to continue transmitting new segments, maintaining transmission in flight while avoiding a full reset. For each additional duplicate ACK, cwnd is increased by one MSS, which permits the transmission of new segments and maintains pipeline fullness. When a non-duplicate ACK arrives that acknowledges new data, cwnd is then set to ssthresh and the connection moves to congestion avoidance. If a timeout occurs, the algorithm reverts to slow start, setting cwnd to MSS and resetting ssthresh accordingly.

History and variants: Fast Recovery was introduced in the TCP Reno era as an improvement over the Tahoe variant, which would reduce cwnd to one MSS on loss. NewReno later refined Fast Recovery to handle certain patterns of multiple packet loss without a complete reset. Many modern TCP stacks implement variants or replacements, such as CUBIC or BBR, but Fast Recovery remains part of the historical foundation of TCP congestion control.

Limitations: Fast Recovery assumes relatively isolated losses; performance degrades with multiple losses within a window. In high-bandwidth, long-distance networks, newer algorithms may yield better throughput and fairness. It remains a standard reference in textbooks and RFCs, notably RFC 2001, which documents fast retransmit and fast recovery.