[e2e] Random losses on very high speed networks

[Cannara]

I'll make one clarifying comment here on Shannon's theorem, as Lloyd states
it.  C is the theoretical channel capacity.  It is not actual.  This is
extremely important for anyone in the business to understand.  The way C is
approached is by choice of symbols to send.  For instance, FEC is a way in
which C can be approached under some noise constraints.  So is the way in
which 'simple' modem protocols have been developed over the years, to deal
with the 4kHz local-loop channel provided by telcos.  

Shannon did not fully specify how symbols were to be designed to allow C to be
approached, nor did he mean that bits were being sent as bits.  Binary
representation of the data was mathematically convenient.  In real, physical
links, signalling symbols can be complex, multilevel, self-correcting, etc.
The external digital interfaces may send/receive bits, but the heavy lifting
is not done that way on the link, except where signalling media are still
primitive, as in optical laser transceivers.

Alex

Lloyd Wood wrote:
> 
> On Tue, 22 Jul 2003, Constantine Dovrolis wrote:
> 
> > Besides, the context in which the term is used is so different
> > for Hz vs bps that probably it never causes confusion.
> 
> Apart from something like Shannon's law, which uses both bandwidth and
> capacity, where the context requires that both be used correctly.
> 
> > And at the end of the day, the two "bandwidth" concepts are not
> > totally irrelevant after all - see Shannon's law.
> 
> That's rather the whole point.
> 
> C = W log2(1+S/N)
> 
> where C is the channel capacity in bits/second, and W is the bandwidth
> in Hz. Notice how C != W. Shannon was very very clear on the
> distinction between bandwidth and available capacity.
> 
[clip]