[e2e] Why do we need TCP flow control (rwnd)?

John Day day at std.com
Fri Jul 11 10:49:48 PDT 2008


I agree with Christian that your second point has more validity.

While there may have been members of the old guard who believed that 
eventually data networks would look like phone networks, it was far 
from a common idea among those thinking about the problem.  A primary 
difference in the argument for data networks at the time was that 
they were phone networks weren't.

As I said some weeks ago, I remember being aware that Poisson wasn't 
a good approximation in the mid-70s and initiating a literature 
search to find alternatives approaches.  I remember turning up things 
like diffusion approximation method, but never had a chance to pursue 
it.

I am afraid that this one more thing we have to chalk up to "grade 
inflation" as the field expanded.  It is just too bad that it spills 
over into textbooks.



At 8:54 -0700 2008/07/11, Christian Huitema wrote:
>  > The reason folks thought that data networks would be Poisson was that
>>
>>      1. Phone networks were demonstrably Poisson
>>
>>      2. They had nothing else in the arsenal (that is, if you were going
>>         to do anything non-trivial, you lacked an analytic alternative)
>
>Craig,
>
>The point about nothing else is certainly valid. If you have to 
>perform your computation using log tables and an actual spreadsheet, 
>then tackling anything besides Poisson will be very challenging. 
>However, it has been known for some time that Poisson was at best a 
>first degree approximation, even for phone traffic.
>
>Consider two widely documented cases, time variations in traffic and 
>flash crowd, both of which have been part of traffic engineering for 
>a long time. The average traffic load is known to vary as a function 
>of the time of day, day of the week, month of the year, not to 
>mention Mother's day. There is no way you can model that as a single 
>"independent arrival" process. Flash crowds happen when many people 
>decide to pick up their phone at the same time, e.g. to call in a TV 
>show, or to report about a particular incident. Again, there is no 
>way to model flash crowds as independent arrivals. Even the duration 
>of phone calls was probably not well modeled by time independent 
>departure processes, e.g. if you take into account the behavior of 
>teenagers in the 70's.
>
>I suspect that any process that results from human behavior is very 
>likely to exhibit much more variability than predicted by a Poisson 
>model. That's certainly true for phone traffic as well.
>
>-- Christian Huitema



More information about the end2end-interest mailing list