Re: [8023-POEP] Claasification Ad hoc - Enhanced classification scheme
Christian and Charles,
You both bring up good points, and I would like to respond.
Here is a brief summary of some of the key points discussed in the
classification adhoc, as far as I understand the general feeling of the
group:
1. The only way to really improve efficiency of the system is by advanced
power management, and that has to be done with a layer-2 protocol. L2
allows dynamic negotiation and re-allocation. That's a big discussion, and
I don't want to get too deep into it here, for the sake of brevity. I only
point this out to provide context for point 2 below.
2. We still have to improve the L1 clasification protocol, but we don't
want to go to heroic extremes; just some reasonable improvement in
granularity that won't add much cost. Cost would include the added
complexity in the hardware and software, and any extra test cases required
to verify the PSE ability to distinguish between all the class signatures.
3. So, given that we're leaning towards including a L2 protocol (that would
presumably cover vendor ID, device model, etc.) and we want to keep the L1
protocol as simple as possible, I would say Charles ideas probably won't
fly.
4. Now regarding Christians ideas about coding: The ping-pong method does
seem to be the front-runner because of it's flexibility compared with
time-based encoding. But the preliminary consensus of the adhoc group seems
to be that we need only about 30 classes. (Although this is still being
debated.) It seems the ping-pong method, even with just three pulses, has a
surplus of extra codes. Another concern, is the potential thermal issues
associated with the class 4 signature currents.
So it appears the best coding scheme would yield just enough codes, while
minimizing heating in the chips.
Therefore, I would propose the following ping-pong coding approach:
a. Assume 3 pulses. Let the levels be only 1, 2, and 3.
No class 4 levels for the moment. This yields 3^3=27 codes.
b. To avoid mistaking an Af-PD for an At-PD, we exclude codes
{1,1,1}, {2,2,2}, and {3,3,3}. This leaves 27-3=24 codes.
c. 24 codes is a little bit light by many peoples opinions. So
now we start adding some that use the class 4 level, but only
until we have about 30 codes, and starting with the codes that
produce the least amount of heating in the chips: {4,1,1},
{1,4,1}, {1,1,4}, {4,2,1}, {4,1,2}, and {4,2,2}. That brings
us up to 30 codes.
d. The 32 remaining code combinations could be reserved for
"future" use.
I hope this was helpful. Again, I don't speak for the adhoc group; the
above comments reflect my opinions and understanding of the groups general
feelings, but I could be wrong.
Steve Robbins
-----Original Message-----
From: owner-stds-802-3-poep@xxxxxxxx [mailto:owner-stds-802-3-poep@xxxxxxxx]
On Behalf Of Charles Palmer
Sent: Friday, February 24, 2006 3:08 AM
To: STDS-802-3-POEP@xxxxxxxxxxxxxxxxx
Subject: Re: [8023-POEP] Claasification Ad hoc - Enhanced classification
scheme
Folks
If you are considering multi-cycle classification schemes, then is there any
merit in extending this so that the PD can return additional information
that
identifies it? Such as a Vendor ID and Device ID?
If the data was returned LSB first (that is, Device ID bits 0 and 1) then
such a
scheme could be compatible with Christian's proposed ping-pong scheme
(device ID
LS bits denote power rating) - with the subsequent bits of Device ID and
Vendor
ID being optional.
If the PD returns 2 bits per cycle then 16 cycles (or 17 to account for the
initial af classification) would return a 16-bit Vendor ID and 16-bit Device
ID.
Maybe that could help the PSE decide what type of PDs should get power and
who
should not, and allow for non-intelligent applications (such as LED
lighting).
Regards - Charles Palmer
Christian BEIA wrote:
> Hi all,
>
> I prepered a short presentation about classification scheme.
>
> Christian