RFC 6582, "The NewReno Modification to TCP's Fast Recovery Algorithm", April 2012

Errata ID: 6871
Status: Verified
Type: Technical
Publication Format(s) : TEXT
Reported By: Clive Bloom
Date Reported: 2022-03-07
Verifier Name: Martin Duke
Date Verified: 2024-03-12

Section 4.1 says:

If the Cumulative Acknowledgment field didn’t cover more than
recover, check to see if the congestion window is greater than
SMSS bytes and the difference between highest_ack and
prev_highest_ack is at most 4*SMSS bytes. If true, duplicate
ACKs indicate a lost segment (enter fast retransmit).
Otherwise, duplicate ACKs likely result from unnecessary
retransmissions (do not enter fast retransmit).

It should say:

If the Cumulative Acknowledgment field didn’t cover more than
recover, check to see if the congestion window is greater than
SMSS bytes and the difference between highest_ack and
prev_highest_ack is at most 3*SMSS bytes. If true, duplicate
ACKs indicate a lost segment (enter fast retransmit).
Otherwise, duplicate ACKs likely result from unnecessary
retransmissions (do not enter fast retransmit).

Notes:

RFC6582 (as well as RFC3782) references to Gur03 and GF04 papers as to the initial sources
of the heuristics both ACK-based and Timestamp-based. Neither of those
papers nor Gur03 nor GF04 defines difference between highest_ack and previous_highest_ack
of at least 4*SMSS bytes upon receiving the third duplicate ACK as an indication
of droped retransmitted segment. Instead, section III of GF04 says:

"The acknowledgment heuristic is based on an observation that if the
TCP sender unnecessarily retransmits at least three adjacent packets,
there will be a jump by at least four segments in a cumulative
acknowledgment field. The sender will have correctly retransmitted at least
one packet, to advance the cumulative acknowledgment field, and
unnecessarily retransmitted at least three more to result in three duplicate
acknowledgments. Following the advancement of the cumulative acknowledgment
field, the sender stores the value of the previous cumulative acknowledgment
as prev_highest_ack and stores the latest cumulative acknowledgment as
highest_ack. Upon receiving the third duplicate acknowledgment,
the sender invokes a Fast Retransmit if its congestion window is greater
than one MSS (Maximum Segment Size), and the difference between highest_ack
and prev_highest_ack is at most three MSS."

According to GF04 if TCP sender in absence of any droped acknowledgments upon receiving
the third duplicate ACK has difference between highest_ack and prev_highest ack values
of at most/i.e. no more than 3*SMSS bytes then this is explicite indication of droped retransmitted
segment and leads TCP sender to invoke Fast Retransmit, but current description of ACK-based
heuristic in RFC6582 section 4.1 in part of: "If the Cumulative Acknowledgment field didn’t cover more than recover, check to see if the congestion window is greater than
SMSS bytes and the difference between highest_ack and prev_highest_ack is at most 4*SMSS bytes.
If true, duplicate ACKs indicate a lost segment (enter fast retransmit). Otherwise, duplicate ACKs
likely result from unnecessary retransmissions (do not enter fast retransmit). ", makes TCP sender
to treat difference between highest_ack and prev_highest_ack of 4SMSS bytes upon receiving 3rd
duplicate ACK as indication of lost retransmitted segment but again according to GF04 this is NOT so,
and makes TCP sender to invoke Fast Retransmit when in fact those three duplicate acknowledgments
indicate unnecessarily retransmitted segments and have in their acknowledgment fields sequence
number which receiver expects to receive next but which sender has NOT sent yet, so Fast Retransmit
has no point in this case.

Report New Errata