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

[Lachlan Andrew]

2008/7/11 Craig Partridge <craig at aland.bbn.com>:
>
> You achieve research by looking at the points more carefully.
>
> * Analysis is useful when employed correctly.

Exactly.

>  For instance, for all
>  that Poisson is terrible, if you can prove a negative result with
>  Poisson arrivals (e.g. service model Zed is a disaster even if
>  arrivals are Poisson) that's a very strong result (as we know
>  Poisson is nicer than real traffic -- per Christian's note).

Rules of thumb like "Poisson is nicer" also have to be treated with
caution.  As David Reed pointed out, simple Poisson models don't model
the fact that users back off during congestion.  In that sense,
Poisson is less nice than real traffic.

There's no substitute for checking how the specific approximations of
a model affect the specific conclusions you are drawing.

> Or, to make this short -- we make progress by being (a) careful and (b)
> smart (in that order).

As long as "careful" comes before "publish, standardise or deploy", I
don't think it always needs to come before "smart".  It is good not to
be too constrained when trying to make progress.  The inventors of
turbo codes admit they wouldn't have thought of their ideas if they
had been careful about the information theory behind them.

Cheers,
Lachlan

-- 
Lachlan Andrew Dept of Computer Science, Caltech
1200 E California Blvd, Mail Code 256-80, Pasadena CA 91125, USA
Ph: +1 (626) 395-8820 Fax: +1 (626) 568-3603
http://netlab.caltech.edu/lachlan