Re: [BP] Relationship to cabling standards
Brian,
Interesting points. One big difference to also
remember is that 802.3 doesn't write cabling standards. I've been
under the impression that .3ap is not going to write a backplane specification
(materials, connectors, etc.) either. If that's true, then it is fair for
the Task Force to assume that someone well-versed in backplane technologies can
use the channel models in the standard to create any implementation that they
desire. For example, a specified channel model with a set of parameters
for IL, RL, NEXT, FEXT, etc. permits implementers to design backplanes with
different materials and layouts to achieve various reaches and cost
requirements. A PHY designed to operate over the defined channel should
work on each implementation that satisfies the channel
model.
I think the question comes back to what should that
channel model be. If there is a channel model for 1G, then why doesn't the
Task Force use that model as a starting point? It appears that the reason
some do not wish to use that model is that it constrains some of the choices of
a PHY device. Does the selection of a channel model impact the broad
market potential? That will be question that the Task Force will have to
be able to answer when they get to Working Group and Sponsor Ballot. If
the channel model for 10G prevents the use of any 1G or XAUI backplanes, then
some may believe that limits the market potential.
Cheers,
Brad
I believe there are important differences between the emerging
802.3ap standard and existing cabling standards.
- Cable, by its physical nature, does not have the potential to have
stubs.
- Cable manufacturers are further prevented from creating ill-behaving
electrical channels via the existence of cabling standards.
- The variation in electrical characteristics of cables will be very small
compared to the electrical variations in the universe of backplanes.
Consequently, the problem space for cable channels is very narrow while the
problem space for backplane channels is very wide, and has more dimensions.
- A particular implementation is not difficult to design signaling for,
it's the huge universe of bad and weird channels that causes the problem for
signaling. Each one may represent 1% of the market, but putting an
operating envelope around ALL of them is what is going to cost the industry
in terms of die area, power, time to market and widescale availability.
- We are expecting to see the number of 802.3ap transceivers-per-IC
sometimes appoach 100 transceivers. The "costs" mentioned above will
be multiplied by this number. What might have been done for something
like 1000BaseT, where a couple transceivers/IC might be seen, is not
appropriate here.
BrianS