Delivery-Date: Thu Dec 11 10:21:02 2008
Message-Id: <200812111823.mBBINWas029887@bourbon.usc.edu>
To: cs551@merlot.usc.edu
Subject: Re: DupACK Spoofing
Date: Thu, 11 Dec 2008 10:23:32 -0800
From: Bill Cheng <william@bourbon.usc.edu>

Someone wrote:

  > I don't think I'm completely understanding the reasoning or benefit behind
  > DupACK Spoofing. So I understand the following happens after a triple
  > DupACK: cwnd = cwnd/2 + 3*SMSS. After this, each additional DupACK causes:
  > cwnd = cwnd + 1. So this should cause the sender to send one new packet. I
  > assume then that the receiver would continue to send additional DupACK's,
  > each one resulting in one new packet of data being sent by the sender.
  > However, this should only be able to continue until the sender's cwnd
  > reaches the receiver's advertised window. At this point, to receive any more
  > data from the sender, the receiver must send a new ACK, which would reset
  > the sender's cwnd to ssthresh, which is half of the original cwnd (before
  > the DupACK's started) due to Fast Recovery.
  > 
  > So really the only benefit was a momentary increase in bandwidth to the
  > receiver until the sender's cwnd is reset, and at that point the cwnd is
  > only half of what it originally was. Maybe the receiver could benefit more
  > by advertising a very large window so that the sender simply continues to
  > increase its cwnd?
  > 
  > Does any of this sound about right? :)

We didn't really go over any details about this in class.

Usually, DupACK Spoofing is not done during fast recovery.
Imagine if you are at the beginning of slow start.  When you
send out the first packet, you get 20 ACKs for this packet.
You will retransmit the first packet, but you will also
inflate the window size when you are not suppose to.
--
Bill Cheng // bill.cheng@usc.edu <URL:http://merlot.usc.edu/william/usc/>