[Iccrg] Re: Appropriate rate given corruption

[Lachlan Andrew]

Greetings Shiv,

On 28/08/07, Shivkumar Kalyanaraman <shivkuma at ecse.rpi.edu> wrote:
> So if the erasure rate of a channel with nominal bit
> rate C is p, then the shannon capacity of that erasure channel is C(1-p).
>
> Now think of a network with new links (indexed by j) where the link
> capacities are replaced with Cj(1-pj). Kelly's theory can then be applied
> to this revised network to define a notion of fairness.

That applies if there is per-hop erasure correction.  In the case
where TCP is exposed to the erasures, it is a little more complicated
than that.
a) We need to model the thinning of downstream traffic, as well as the
reduction in capacity of the lossy link
b) Kelly's framework doesn't distinguish between the "sent rate" and
the "received rate".  With significant loss, we need to consider the
fact that utility is only derived from the received rate.
c) Most lossy links are wireless.  In the common case that the
wireless link is a "bus", the loss rate is typically different for
different receivers (i.e., different flows).  That is why I modelled
the loss rate as a property of the flow at the link, rather than just
the link.

> a scheme that offers superior reliability due to better coding (eg: hybrid
> ARQ/FEC) should not be considered unfair compared to a scheme that uses
> all the same congestion detection/response mechanisms but uses a less
> effective reliability strategy (eg: window ARQ/SACK etc).

Absolutely.  Fairness should IMO be measured in terms of how much harm
is caused vs how much benefit is received.  If two flows cause equal
incremental  harm per extra unit of send-rate, then the one which
receives greater  incremental  benefit per extra unit send-rate should
have its rate increased.

Cheers,
Lachlan

-- 
Lachlan Andrew  Dept of Computer Science, Caltech
1200 E California Blvd, Mail Code 256-80, Pasadena CA 91125, USA
Phone: +1 (626) 395-8820    Fax: +1 (626) 568-3603