Delivery-Date: Wed May  2 09:59:01 2007
Message-Id: <200705021658.l42GwAhg024521@bourbon.usc.edu>
To: cs551@merlot.usc.edu
Subject: Re: lecture 16 slide 21 
Date: Wed, 02 May 2007 09:58:10 -0700
From: william@bourbon.usc.edu

Someone wrote:

  > I wanted to say that what we have in the slide indicated
  > as *knee* should be the *cliff*. thats it.

If you are talking about only slide 21 of lecture 16, you are
correct and I have just fixed the slides.

But if you are talking about slide 50 of lecture 12 and slide
4 of lecture 13, then I would still disagree.  According to
what you have mentioned (from page 159 paragraph 2 of
[Ramakrishnan90a]:

    We call the point at which throughput falls off rapidly
    before the network reaches congestion collapse as the cliff.

This matches with what we have on slide 4 of lecture 13:

    http://merlot.usc.edu/cs551-s07/lectures/lecture13/slides-4.html

So, we are consistent with the paper.
--
Bill Cheng // bill.cheng@usc.edu <URL:http://merlot.usc.edu/william/usc/>


  ----- Original Message -----
  From: william@bourbon.usc.edu
  Date: Tuesday, May 1, 2007 9:58 pm
  Subject: Re: lecture 16 slide 21
  To: cs551@merlot.usc.edu
  
  > Someone wrote:
  > 
  >  > Its kind of confusing for me  what u replied back in the last
  >  > mail. Actually its clearly indicated in the research paper by
  >  > Ramakrishnan page no 159 paragraph 2
  >  > 
  >  > "We call the point at which throughput falls
  >  > off rapidly before the network reaches congestion collapse as the
  >  > cliff. This is
  >  > also the point at which the response time approaches infinity.
  >  > The purpose of a
  >  > congestion control scheme [2, 14] is to detect the fact that the
  >  > network has
  >  > reached the cliff, resulting in packet losses, and to reduce the
  >  > load so that the
  >  > network can recover to an uncongested state. "
  > 
  > So far so good...  Like you said, this is about the "cliff".
  > 
  >  > this clearly means that they also mean the cliff and knee as the
  >  > same thing as in congestion control mechanism
  > 
  > I don't know why you said that!  If you read the next two
  > sentenses, it defines the "knee" to be at a different point
  > (same as what we showed in our slides).
  > 
  >  > So i guess there should be no reason for multiple understanding
  >  > of knee and cliff
  > 
  > I would disagree.
  > --
  > Bill Cheng // bill.cheng@usc.edu 
  > <URL:http://merlot.usc.edu/william/usc/>
  > 
  > 
  > 
  >  ----- Original Message -----
  >  From: william@bourbon.usc.edu
  >  Date: Tuesday, May 1, 2007 1:44 pm
  >  Subject: Re: lecture 16 slide 21
  >  To: cs551@merlot.usc.edu
  >  
  >  > Hi,
  >  > 
  >  > Just want to mention that not all people agree on the
  >  > terminology used in [Ramakrishnan90a].  In [Ramakrishnan90a],
  >  > the authors used the term, "knee of the delay curve" to mean
  >  > what others would call the "knee of the throughput curve".
  >  > 
  >  > Others would argue that the "knee of a curve" is the part
  >  > that "bend the most" (i.e., smallest radius of curvature).
  >  > If you look at the delay curve, it bends the most at the
  >  > point that corresponds to the "cliff of the throughput curve".
  >  > --
  >  > Bill Cheng // bill.cheng@usc.edu 
  >  > <URL:http://merlot.usc.edu/william/usc/>
  >  > 
  >  > 
  >  >  -----Original Message-----
  >  >  Date: Mon, 30 Apr 2007 22:00:12 -0700
  >  >  From: william@bourbon.usc.edu
  >  >  To: cs551@merlot.usc.edu
  >  >  Subject: Re: lecture 16 slide 21 
  >  > 
  >  >  Someone wrote:
  >  >  
  >  >    > In lecture 16 slide no 21(power and load) I guess in the 
  > graph  >    > the Knee line is shown wrong and it should be the 
  > first line..
  >  >  
  >  >  You are correct.  There are other slides that show the knee
  >  >  at the right place:
  >  >  
  >  >      slide 50 of lecture 12
  >  >      slide  4 of lecture 13
  >  >  
  >  >  Thanks for catching this one.
  >  >  --
  >  >  Bill Cheng // bill.cheng@usc.edu 
  >  > <URL:http://merlot.usc.edu/william/usc/>