[Tsvwg] a review for draft-ietf-tcpm-tcp-dcr-03.txt

[Sally Floyd <[e-mail]>]

I reviewed this draft, and in terms of technical content I think
it is ready to ship.  I have one suggestion for presentation (a
SHOULD, not a MUST), and that is that the document include an
overview of the algorithm before jumping into the details of Sections
2.1, 2.2, 2.3, and 2.4.  That suggestion, and a few nits, are in
the review below.

- Sally

Feedback for "Improving the Robustness of TCP to Non-Congestion Events",
draft-ietf-tcpm-tcp-dcr-03.txt.

Presentation:

For the example TCP scenario in the Introduction, a figure would
be useful (e.g., as recommended in draft-iab-model-03.txt, at
"http://www.iab.org/documents/drafts/draft-iab-model-03.txt").

The paper also jumps into a detailed description of the algorithm
without giving the reader a good overview of what is going on.
E.g., we are in the middle of Sections 2.1, 2.2, 2.3, and 2.4 before
we have an overview of the algorithm.  A non-normative description
of an example use of the algorithm would be useful in Section 2.0
first (paralleling the example TCP scenario in the Introduction).
Again, this follows the recommendation in draft-iab-model-03.txt
cited above.

As part of this, the paragraph about "LT_F" was confusing to me,
because it wasn't preceded by much context about the overall
algorithm.


Minor editing:

Abstract:
 "In addition, we specify a change to TCP's
 congestion reaction decision point, as well (but, do not require such
 a change to use NCR)."

I found the sentence above confusing.  I figured out later that it
was referring to Aggressive Limited Transmit, not reducing the
sending rate until a packet is actually retransmitted, but it wasn't
clear to me when I first read it.  Perhaps you could just say
something like the following:

 "In addition, we specify an option, Aggressive Limited
 Transmit, where the TCP sender doesn't reduce its sending rate
 until a packet is actually retransmitted; this would delay
 the reduction of the sending rate by roughly one round-trip time,
 compared to current TCP implementations.


P. 4:
"First, the trigger for retransmitting a segment is changed from three
duplicate ACKs [RFC2581,RFC3517] to a congestion window's worth of
duplicate ACKs."

Actually, to at least a congestion window's worth of duplicate ACKs,
right?  (For cwnd < 3, three dup acks are still required.)


Nits (some of these are just my own preferences):

P. 2:
"out of order duplicate ... " -> "out-of-order, duplicate ..."

"segment loss"?  Isn't it "packet loss"?  (Because it is already
a packet when it is lost in the network.)

P. 4:
"needs retransmitted" -> "needs to be retransmitted"

P. 5:
"needs to be retransmitted." -> "needs to be retransmitted" (no period)
"TCP-NCR implementation MUST" -> "A TCP-NCR implementation MUST" ("A")
"reflects" -> "determines"

P. 7:
"back to operating point" -> "back to the operating point"

P. 8:
Change:

"If the following comparison holds ...
...
  (pipe + Skipped) <= (FlightSizePrev - SMSS)"
  
to:

"If the comparison in equation (1) below holds ...
...
  (pipe + Skipped) <= (FlightSizePrev - SMSS)     (1)"

P. 10:
"significant impact" -> "significant impact,"

P. 11:
"the Extended Limited Transmit algorithms" ->
"the Extended Limited Transmit algorithms specified in this document"

"to not be" -> "not to be"

_______________________________________________
tsvwg mailing list
[e-mail]
https://www1.ietf.org/mailman/listinfo/tsvwg