RE: [10GBASE-T] Technical feasibility: power
Pat,
I agree that power requirements and constraints in the use of Ethernet
technologies have changed over the years. I hope that the PHY vendors
will provide estimations of the power requirements relative to selecting
their proposal, and some have already given some rough power estimation
numbers. What if we select 5W as a number for the PHY to meet, but the
first PHY in .13um is 6W? Have we failed the objective? If the next
step in the process technology drops the power to 4W, can we consider
that we met the objective with the first 6W PHY? It's a tough number to
target, and most of the numbers we're using are guestimates. I think
that the TASK FORCE (not the Study Group) should take into account the
power when selecting proposals, but do we want to forego the standard
development because we don't know the exact power requirement? That's
the question the study group needs to answer. The feedback I get from
customers is that they do want to see us move forward.
Thanks,
Brad
-----Original Message-----
From: pat_thaler@agilent.com [mailto:pat_thaler@agilent.com]
Sent: Friday, August 01, 2003 12:45 PM
To: Booth, Bradley; xichen@marvell.com; stds-802-3-10gbt@ieee.org
Subject: RE: [10GBASE-T] Technical feasibility: power
Brad,
I'm pretty sure there wasn't an explicit power objective for GbE and
10GbE though power was discussed in 1000BASE-T (I think mostly with
regard to whether it would be low enough to make putting the thing in a
chip feasible).
Needs change with time. For 1000BASE-T initial use in end nodes, one
just had to get it on a board with a gig MAC chip. I don't think there
was any doubt that if the power allowed you to get it into a chip you
could put it onto a board.
The thing that is different is that the 10 Gig server market wants more
than a MAC on the card. We are being asked to put TCP/IP Offload, RDMAP
and iSCSI protocol, and in some cases IPSec on the card. These burn a
lot of power too. That is why I think there should be a power objective
when there hasn't been one in the past.
Here's what I've gotten from the board specs:
PCI Express -
for a 4x or 8x card (this is the backplane bandwidth suitable for 10
Gig), a standard height card is allowed 25 W, a low profile card is
allowed 10 W. The 25 W number assumes that sufficient cooling is
provided by the system.
Infiniband -
for a standard single wide slot is 25 W.
Generally the systems vendors prefer that we stay under 15 W as a system
full of 25 W cards is a lot for them to handle.
If a 10GBASE-T transceiver was under 5 W or so, that would leave us
power for the other functions we need on the PCI Express standard height
card and on Inifiniband. That power level probably wouldn't enable low
profile cards with full offload functions.
Regards,
Pat
-----Original Message-----
From: Booth, Bradley [mailto:bradley.booth@intel.com]
Sent: Thursday, July 31, 2003 2:45 PM
To: xichen@marvell.com; stds-802-3-10gbt@ieee.org
Subject: Re: [10GBASE-T] Technical feasibility: power
I was trying to recall if 802.3 used and specific power numbers in the
GbE and 10GbE objectives or 5 criteria. Considering I don't have web
access right now, I'm just going by memory. I believe 802.3 did not
dictate power requirements, but did use power estimations in selection
of the PHY to put into the draft. This is just a Study Group, so it may
only be required that to state that due consideration will be given to
the power requirements.
Thoughts?
Thanks,
Brad
-----------------
Sent from my BlackBerry.
-----Original Message-----
From: xichen@marvell.com <xichen@marvell.com>
To: stds-802-3-10gbt@ieee.org <stds-802-3-10gbt@ieee.org>
Sent: Thu Jul 31 13:52:04 2003
Subject: RE: [10GBASE-T] Technical feasibility: power
Burce,
That is what I called painful lessons. The customers always want parts
that have high performance margin and low power. I believe that has
become the rule for our designers unless you really want something that
can also fry eggs. In order to give a reasonable estimation of the
power comsuption, we should defintely take into account all possible
technology and design tricks that we know and we can use in 5 years. Of
course, we cannot predict any magic stuff that can save the world will
come out soon, but we can wait.
A well-estimated number will save a lot words.
Xiaopeng
Bruce Tolley <btolley@cisco.com>
07/31/2003 12:56 PM
To: xichen@marvell.com, Bruce Tolley <btolley@cisco.com>
cc:
Subject: RE: [10GBASE-T] Technical feasibility: power
Xiaopeng
Thanks for the response
The first 1000BASE-T parts were much more than a few watts. Also the
power numbers on the roadmaps of the various vendors started dropping as
soon one of the competitors achieved lower power numbers. All the
sudden, the impossible became possible.
And yes we need data to be presented in September.
Bruce
At 11:10 AM 7/31/2003 -0700, xichen@marvell.com wrote:
Hi,
I believe that everyone of us has painfully learned a lot from the
history of 1000BASE-T. We have observed the power of a single
1000BASE-T dropped from a few watts to sub-watt level today. And we
can pretty confidently assess the impact of the technology that we are
gonna use in 5 years and we definitely should take that into account
NOW.
Many circuit design experts will tell you that the power of digital
circuit part can be scaled down while the semiconductor technology got
significant improvement in the following years (how about using 65nm
when 10GB-T reach the market), but the analog circuit part won't get too
much benefit from that and its performance (for example, the ADC
resolution) will be limited by some fundamental physical rule. Due to
the complexity increase of the DSP part and the analog part, even using
65nm technology for digital circuit and using SiGe technology for analog
circuit, to reach 100m on CAT-7, the estimated power of the transceiver
(assume it is practically feasible) will be a number that can surprise
you. I believe more and more data will be given in the following
meeting to show you the reality.
Xiaopeng
Bruce Tolley <btolley@cisco.com>
Sent by: owner-stds-802-3-10gbt@majordomo.ieee.org
07/31/2003 10:12 AM
To: pat_thaler@agilent.com, yousefi@broadcom.com,
pat_thaler@agilent.com, btolley@cisco.com, bradley.booth@intel.com,
stds-802-3-10gbt@ieee.org
cc:
Subject: RE: [10GBASE-T] September interim meeting
Pat
Thanks for providing detail on data centers. I would argue that in terms
of broad market potential, 10GBASE-T would pass muster even if the only
market application was data centers.
On the power issue, the first 1000BASE-T implementations did not appear
until well after the standard was done, some 5 years after the High
Speed Study Group got its PAR, and consumed an obscene about of power.
We might have never achieved the low power 1000BASE-T PHYs we have
today if we had tried to agree on exact numbers in 1996.
Bruce
At 11:00 AM 7/31/2003 -0600, pat_thaler@agilent.com wrote:
Nariman,
CX4 is useful especially when we have in rack connections to make or
ones going to the next rack. However, the distance is too short for many
other data center connects. Also, the cable for the long distance is
relatively bulky which may be a problem for some uses. We will be glad
to get it, but it only solves a corner of the problem space.
Something for the longer distances in data centers that is lower cost
than fiber would be useful. For that environment, it doesn't necessarily
have to rely on already installed wiring. Running on existing wiring is
nice, but not essential.
My view of the important items for the data center environment:
It must perform solidly on the media we choose for it - data integrity
factors such as BER must be met.
It must be able to live on "standard" server bus adapter formats with a
TOE: e.g. PCI Express and Infiniband
which means power is a concern
It must be transparent to existing MACs - that is, the MAC must see the
same behavior it sees with 10 Gig fiber.
100 m would be desireable (partly to enable future horizontal usage) but
the data center could live with shaving something off that. (100 m is
nice from a standards development standpoint as it saves us from arguing
about what lower number is enough.)
The media it runs over should not be so stiff or bulky that it is a
problem to accomodate with normal rack and data center cable management.
Of couse it must also meet EMI requirements
Regards,
Pat
-----Original Message-----
From: Nariman Yousefi [ <mailto:yousefi@broadcom.com>
mailto:yousefi@broadcom.com]
Sent: Wednesday, July 30, 2003 8:10 PM
To: pat_thaler@agilent.com; btolley@cisco.com; bradley.booth@intel.com;
stds-802-3-10gbt@ieee.org
Subject: RE: [10GBASE-T] September interim meeting
Pat,
I agree that the issues you raised must be addressed by November. One of
the biggest challenges for this group is to establish reality on
technical feasibility on Cat7, Cat6 and Cat5e channels. Different
vendors have different conclusion on Technical feasibility. That is due
to assumptions on alien cross talk mitigation techniques, impact on
implementation impairments on SNR, channel model, coding gain, and
analysis on chip complexity in a given process. Assumptions must be
stated clearly by vendors that present technical feasibility. In this
case, technical feasibility drives the broad market potential.
Technical feasibility must be addressed at least based on the following
criteria:
1. Achievable distance on Class D channel with and without installation
mitigation techniques.
2. Achievable distance on Class E channel with and without installation
mitigation techniques.
3. Transceiver complexity in terms of estimated power dissipation and
realistic targets for building blocks like ADC, PLL and etc 2-3 years
from now.
We reached a conclusion that cat7 cable or class F channel has high
enough capacity for 10Gbps operation. But, can a transceiver be built
with reasonable power dissipation and cost say in 90nm process or finer
geometries to achieve broad market potential?
We need to keep in mind that customers have fiber and CX4 as
alternatives.
Nariman
At 01:08 PM 7/30/2003 -0600, pat_thaler@agilent.com wrote:
Bruce,
Generally, when the group can agree on clear objectives, then they can
finish the rest of the work. Fuzzy objectives often indicate a lack of
real concensus.
In November, I will also be expecting arguments that support the 5
criteria based on the objectives -
especially:
Broad market potential - evidence that there will be a broad market the
minimum requirements of the objectives are met.
Technical feasibility - is it feasible to meet those minimum
requirements
Economic feasibility - when you have met the minimum requirements will
cost be suitable to make it a viable product in the markets?
In the discussions at the plenary, a power consumption issue was raised
by some of the speakers.
If the broad market potential is based in part on use in devices such as
end nodes (including servers in data centers), then an objective for
power consumption such that this can reside in server card formats would
be important. Can it fit within the power constraints of a PCI Express
board and an Infiniband board (remembering that one has to allow some
power for the MAC and probably TOE/RDMAP engine)?
Looking at the objectifves in agenda_1_07_03, I don't see any that
address power consumption or the abilitiy to live on server card
formats. In a quick search, I also didn't find any material on power
consumption in the presentations that have been made to the study group.
I hope that in September the group will address the issue of power.
Regards,
Pat
-----Original Message-----
From: Bruce Tolley [ <mailto:btolley@cisco.com>
mailto:btolley@cisco.com]
Sent: Monday, July 28, 2003 1:22 PM
To: Booth, Bradley; stds-802-3-10gbt@ieee.org
Subject: Re: [10GBASE-T] September interim meeting
Brad:
Thanks for the follow up.
I am confident that if we can agree on crisp, clear objectives for 10
Gbps reach and media supported in September that we can get our PAR
approved and move into Task Force mode, which is where the real work
begins.
Bruce
At 06:35 PM 7/24/2003 -0700, Booth, Bradley wrote:
Study Group Members,
Just to let others that were not at the meeting know the outcome of the
802.3 Working Group meeting, the Study Group will have to complete its
PAR, 5 Criteria and Objectives in November. This gives the Study Group
the task of completing the PAR, 5 Criteria and Objectives in 4 months.
This will make our September Interim meeting extremely important. We
will need to complete the effort as much as possible to pre-submit to
the 802.3 Working Group prior to the November Plenary. November will
permit us the ability to modify the PAR, 5 Criteria and Objectives prior
to asking 802.3 to put the PAR on the NesCom agenda. The September
Interim meeting will focus on the completion of our PAR, 5 Criteria and
Objectives.
Thanks,
Brad
Chair, 10GBASE-T Study Group
Bruce Tolley
Senior Manager, Emerging Technologies
Gigabit Systems Business Unit
Cisco Systems
170 West Tasman Drive
MS SJ B2
San Jose, CA 95134-1706
internet: btolley@cisco.com
ip phone: 408-526-4534
"Don't put your hiking boots in the oven unless you plan on eating
them."
Colin Fletcher, The Complete Walker
Nariman Yousefi
Vice President Networking Engineering
PH (949) 585 5450
FAX (949) 453 1848
e-mail : Yousefi@Broadcom.com
Bruce Tolley Senior Manager, Emerging Technologies Gigabit Systems
Business Unit Cisco Systems 170 West Tasman Drive MS SJ B2 San Jose, CA
95134-1706 internet: btolley@cisco.com ip phone: 408-526-4534
"Don't put your hiking boots in the oven unless you plan on eating
them."
Colin Fletcher, The Complete Walker
Bruce Tolley Senior Manager, Emerging Technologies Gigabit Systems
Business Unit Cisco Systems 170 West Tasman Drive MS SJ B2 San Jose, CA
95134-1706 internet: btolley@cisco.com ip phone: 408-526-4534
"Don't put your hiking boots in the oven unless you plan on eating
them."
Colin Fletcher, The Complete Walker