Re: <draft-stevens-tcpca-spec-01.txt>

Fengmin Gong <fmg@crusher.mcnc.org> Thu, 21 March 1996 02:59 UTC

Sender: ietf-archive-request@IETF.CNRI.Reston.VA.US
From: Fengmin Gong <fmg@crusher.mcnc.org>
Message-Id: <9603210250.AA11989@crusher.mcnc.org>
Subject: Re: <draft-stevens-tcpca-spec-01.txt>
To: ietf@CNRI.Reston.VA.US
Date: Wed, 20 Mar 1996 21:50:36 -0500
Cc: Fengmin Gong <fmg@ncren.net>, end2end-interest@isi.edu
Mime-Version: 1.0
Content-Type: text/plain; charset="US-ASCII"
Content-Transfer-Encoding: 7bit
Source-Info: From (or Sender) name not authenticated.

Hi,

I have two comments/questions on the draft.  The first one is on the
Congestion Avoidance and the second on Fast Recovery.

>   3.  When congestion occurs (indicated by a timeout or the reception
>       of duplicate ACKs), one-half of the current window size (the
>       minimum of cwnd and the receiver's advertised window, but at
>       least two segments) is saved in ssthresh.  Additionally, if the
>       congestion is indicated by a timeout, cwnd is set to one segment
>       (i.e., slow start).
This paragraph does not say anything about changing the current cwnd in
the case of duplicate ACKs.  One would be left to assume that cwnd remains
unchanged in this case, which is not consistent with Van Jacobson's
description of Congestion Avoidance behavior.  The consistent behavior
is setting both cwnd and ssthresh to be half the current cwnd
value, which should be the desirable behavior.

>   1.  When the third duplicate ACK in a row is received, set ssthresh
>       to one-half the current congestion window, cwnd, but no less
>       than two segments.  Retransmit the missing segment.  Set cwnd to
>       ssthresh plus 3 times the segment size.  This inflates the
>       congestion window by the number of segments that have left the
>       network and which the other end has cached.

I have a little bit of concern with the twist of inflating the
congestion window by 3 times the segment size.  The reason is that when
cwnd=6 on the third duplicate ACK, halving the cwnd does not have the
real effect of reducing the effective rate of the sender.  Although I do
not have any data to show the negative impact of this, skipping the
inflation step seems to be a much safer thing to do.  Is there any
experiment data with a TCP implementation using the inflation-by-3-seg
modification?  Also In her paper "TCP and Explicit Congestion
Notification," Sally Floyd mentioned that TCP Reno's Fast Recovery
algorithm actually wait half a round trip time and halves the congestion
window.  Has anyone done a comparison study between the wait and no-wait
Recovery algorithm?

Thanks,
--
Fengmin Gong, Network Research Engineer E-mail: gong@mcnc.org
MCNC Information Technologies		Phone:  919 248-9214
3021 Cornwallis Rd., RTP, NC 27709	FAX:    919 248-1455
http://www.mcnc.org/HTML/ITD/ANT/HPCCResearch.html

Re: <draft-stevens-tcpca-spec-01.txt> Fengmin Gong