RE: Signal vs. Idle debate (was: Here's a new idea)
Mr. Polk,
I will try to address your points consisely.
-----Original Message-----
From: James M. Polk [mailto:jmpolk@xxxxxxxxx]
Sent: Friday, May 05, 2000 11:20 AM
To: DOVE,DANIEL J (HP-Roseville,ex1); stds-802-3-pwrviamdi@xxxxxxxx
Subject: RE: Signal vs. Idle debate (was: Here's a new idea)
Dan (and all)
Comments inserted
At 03:39 PM 5/4/2000 -0600, DOVE,DANIEL J (HP-Roseville,ex1) wrote:
>
>Hi Norman,
>
>I have a few issues with the direction you are taking the
>discussion so far. Let me explain.
>
>> If the market is growing very rapidly, then existing owners
>> of (what will soon
>> be) absurdly bulky and expensive switches will be small
>> fraction of the total
>> Ethernet market, and catering to those old switches with
>> add-on boxes to
>> supply power will be a niche market.
>
>On the other hand, we are in the business of selling products
>today to solve problems, not create them. There are plenty of
>switches being sold today that support VoIP but don't supply
>power because there is no standard for it. Are these boxes
>"old, absurdly bulky or expensive"? I don't think so. Some of
>them are highly integrated and very cost effective. I could
>give you examples but I think you know I am right.
Dan -- let's get a little realistic, here. We're not merely talking about
software to be downloaded to existing equipment to take advantage of this
new capability. DTE power means that customer has embraced a new technology.
That means new equipment to buy (in this case). Period. That is a fact.
OK. A customer can buy an HP4000 today that does not provide
power but supports VoIP. They can buy a power-insertion patch
panel (when the standard is defined) and a phone, and they are
now fully operational without ripping out their switch. Is that
clear? This model works for many other switches that are in the installed
base today.
Devices that expect power on either pair are just being built -- therefore
just bought -- therefore nothing is old or existing regarding this coming
Standard. If the switch manufacturers cannot build a new switch module with
power on that module -- shame on them.
The question is not "IF they can do it", it is "should they be
forced to do it"? The committee decision to require mid-span support says
"no".
Companies are already doing it and those that don't won't sell too much
product. Period. This is a fact.
I disagree. I believe that those who apply power at the switch are
burdening the cost of the switch for a limited number of applications. Those
who rely on a patch panel power distributor are actually allowing the
customer to determine which ports to provide power to and keeping their
switch investment optimized.
Isn't this element of the debate moot? The committee has already decided
that mid-span insertion is a "must".
As a point of reference towards past work the IEEE did: 802.1p/q made
everyone purchase new Ethernet switches because no one had built their
existing Switches with the ASIC capable of those additional bytes to the
Ethernet Header. This too is a fact. Period.
But people ARE providing VoIP capability in their current products and their
customers should be able to take advantage of that capability at the least
cost.
So your argument that the IEEE should (basically) only ever adopt what
doesn't impact the existing installed base is a fairly foolish
justification, IMO. And as far as cost of implementation is concerned --
this one cost customers a lot. Now -- I'm personally glad the IEEE did do
this work, there are great benefits to it.
Pardon my foolish concern for my customers, I guess we will have to disagree
about their best interests. I recommend that you have a nice cup of tea and
relax. We can debate this matter without getting to the point of calling
each other foolish or unrealistic.
Back to this topic, what will be engineered is power into existing Ethernet
Switching Chassis with new modules purchased by that new technology
customer.
I have no dispute with those who wish to offer that as a solution. I am
arguing that we should not impair the data pairs for 10/100T or REQUIRE the
customers to toss out their very efficient and cost-effective VoIP capable
switches. Standards should be designed to provide flexibility for the
customer.
I haven't done the numbers -- but for those that have a small fixed Ethernet
Switch in a closet (<50 ports) -- I'd be surprised if building a cost
effective external 'mid-span' power insertion device for those installations
were the norm, not the exception. Companies will build new 'stackable' or
'fixed' or 'non-modular' chassis that will supply DTE power. If you don't
believe that's true -- wait and see. I'll bet there will be many many
offerings by many vendors by the time this 'Standard' is ratified.
If you wish to perform the numbers regarding the justification
for mid-span versus switch-based power insertion, I would be very happy to
review and comment on your presentation.
>> Whether we choose the data pair or the
>> spare pair should be conditioned more on 1) the total cost
>> to the buyer of
>> equipment designed with DTE power in mind, and
>
>As a standards committee member, I would put "must support the
>installed base of equipment without interference" as a higher
>priority than cost. The cost of ripping out existing gear is
>substantial. But what I have to say is less relevant than the
>decision by the committee to require compatibility with 10/100T.
Again, explain 802.1p/q mentioned above
But it does not apply in this case because there is a very large installed
base of VoIP product in the market.
>
>> 2) on the
>> needs of 1G and
>> faster links, than on the cost of existing installations.
>
>Again, the committee has clearly expressed that 10/100T is a
>higher priority than 1000T. Unlike 10/100T, the Gigabit stuff
>today actually operates very robustly. Much of the 10/100T
>installed base has been built with less-than-ideal performance
>but customers want to use it. If you have a PC with an older 100T
>NIC installed, do you want its performance to be impacted by the
>addition of mid-span patch panels that are designed to inject
>power on PDTEs?
I guess I don't understand why a PC would be connected to a Mid-Span
powering device and expecting power --- wouldn't that blow-up that older
NIC? I sure think that NIC couldn't handle -48Vdc on either pair.... but I
could be wrong.
That older NIC would likely fail the "assert power" test, and would
therefore not have power asserted, but might not operate reliably due to
Return Loss and Insertion Loss impacts on the data pairs if they were used
to insert power. If you wish to understand this point more clearly, I
recommend that you review the numerous presentations already given to the
committee. They clearly explain it.
In the case of the "spare pairs" no impact would
>exist. Sure, future 1000T products would have to deal with power
>injection on their outside pairs, but 1000T is better equipped
>to do so with echo cancellers, FEC, and DSP capabilities.
>
>> (Note that this
>> argument says that the woes of existing 4-wire installations
>> is relatively
>> unimportant.) In other words, I think that powering 1G is
>> more important
>> in the long term than cheaply adding on to 10M ports' capabilities.
>
>Unfortunately, that is not the direction that the committee has
>already committed to. Support for 1000T should not be demanded at
>the expense of either reliable or inexpensive 10/100T operation.
>
>> -- Norm
>
>Best Regards,
>
>Dan Dove
>HP ProCurve Networks
>
*************************************
"At the end of the day... the most committed win!"
James M. Polk
Sr. Product Manager, Multiservice Architecture and Standards
Enterprise Voice Business Unit
Cisco Systems
Dallas, Texas
w) 972.813.5208
f) 972.813.5280
www.cisco.com <http://www.cisco.com/>
Best Regards,
Dan Dove
HP ProCurve Networks