Re: [BP] Question regarding Channels
Mike,
Thank you for clarifying your position on some of the issues. It is
interesting that I totally disagree with you on several issues, while
agreeing with you on others. Furthermore, you were one of the critics that
voiced opposition to the channels I proposed, based on guidance from the
signaling ad hoc, and I requested you specifically call out the channels and
the issues with those channels that you had issues with. Please address
this request.
Now as far as other issues in your email- Do I support ATCA - obviously the
answer is yes, but as you pointed out ATCA is a growing market, but not
necessarily the only one in town. However, a backplane is a backplane, and
the issues that are being faced in ATCA are not dissimilar as to what the
entire market is facing. From my perspective, I have been forced by critics
to use the ATCA backplane. Why? Very simple - data that I and others have
presented in the past has been criticized for being not complete, as the
system tradeoffs that need to be made in a real design have not been
included, as they were test vehicles. Using the ATCA backplanes has allowed
me to deflect this critique now. This is exactly the point that Brad was
referring to, using some standard backplane will help us to move forward.
System vendors have not been forthcoming with their own backplane data,
which is understandable.
Now you went through a series of questions, but this is a key one.
"Do we want to support 10 Gigabit Serial Links on Existing ATCA Backplanes
designed for 1 Gigabit or XAUI speeds.
My Answer is NO."
If we were to change the question slightly to
"Do we want to support 10 Gigabit Serial links on existing backplanes
designed for 1 Gigabit or XAUI speeds?"
The answer is totally different. The reality is that the industry has an
installed base of backplanes and implementations, that, if possible, they
would like to try and move to higher speeds. And if they can't, they would
still like to minimize as much of the changes needed in the channel to
support those speeds.
The current informative channel model is based on "improved FR-4" where
minimal stubs have been implemented, and a trace width of 6 mils have been
used on the line cards and backplanes. Note that the last two are
implementation decisions. The requirements of the project only state
operating over 1m of "improved FR-4." Hence, I believe there is a large
number of vendors who would like to see if things can be made to run without
counterboring or increased loss with signaling techniques. You can
interpret the channel model in two fashions - one is the limit of how bad
things can be allowed to be, or second how good can a channel be expected to
get. In the past another group pushed the concept of "Greenfield" channels
that were significantly better than Joel's proposed model. Joel's work
clearly shows using "improved FR-4" and his implementation techniques this
is as good as one should expect things to get, which was significantly
different than this other group's efforts.
And before you throw the power issue up, I will point out that an
alternative technique has been shown, which has shown some initial results
of being able to deal with the stub effect by Mary Mandich, a person who was
most vocal about the power issue.
So with all that stated - let me go through the channels I proposed and see
if i understand your concerns.
Case #1 has slight ripples beneath the SDD21 model with most of it above.
Since this violates the model you reject it?
Case #2 and #3 are the margin cases the Signaling Ad Hoc requested. You
disagree with their use because of violating the SDD21 model, and more power
will be required to make these channels work? So this means you disagree
with the Signaling Ad Hoc's desire to do margin analysis? Every system
vendor I have spoken to agrees with the direction the Signaling Ad Hoc wants
to go, which I also support.
Case #4 and Case #5 you are ok with because they are above the SDD21 model?
Case #6, which is a system 22" top layer connection with the Quadroute
technique, violates the model after 6 GHz with the null. You object to this
because it violates the SDD21 model?
Case #7 violates the SDD21 model @ -55 dB @ approximately 11.5 GHz, you
oppose its inclusion because it violates the SDD21 model?
John
-----Original Message-----
From: Mike-Lerer [mailto:mike@mike-lerer.com]
Sent: Friday, October 08, 2004 9:27 AM
To: STDS-802-3-BLADE@LISTSERV.IEEE.ORG
Cc: DAmbrosia, John F
Subject: RE: [BP] Question regarding Channels
As a System Architect and Designer, I think before we spend overly much time
debating the characteristics of particular channels, we should first see if
we can agree on the problem we are trying to solve.
The group needs to be clear on the problem.
The ATCA market is projected to develop into a large and important market
and as such is of interest to 802.3ap members.
However, today and for the short to medium term future, ATCA is only a small
fraction of the Total Available Market.
The ATCA market will include both Existing (Legacy) and New (Greenfield)
products. Products of potential interest to 802.3ap include:
Line Cards
Switch Cards
Passive backplanes
Do we want 802.3ap to include auto-negotiation to allow support for existing
Gigabit Ethernet and XAUI Line Cards.
My Answer is YES.
Do we want 802.3ap to auto-negotiate to allow a future switch fabric card to
be built that supports existing Gigabit Ethernet Line Cards, XAUI Line Cards
and new 10 Gigabit Serial Line Cards
My Answer is YES.
Do we want to support 10 Gigabit Serial Links on Existing ATCA Backplanes
designed for 1 Gigabit or XAUI speeds.
My Answer is NO.
As a System Designer, I know that there are some parameters that I have
under my control and some I do not.
I have no control over the Mechanical outline, the Thermal Budget, or the
Power Budget for an ATCA card.
I have control over the Line Card and Backplane:
materials and construction
design and manufacturing practices
I desire an optimal solution for the constraints which I am unable to
change:
Space (must integrate between 10 & 100 links on a single ASIC)
Power (because thermal issues are my most severe constraint)
Could the 802.3ap Task Group spend its time seeking a 10 Gigabit Serial
signaling solution that will operate over existing Gigabit and XAUI
backplanes?
My Answer is Yes
Should the 802.3ap Task Group spend its time seeking a signaling solution
that will operate over existing Gigabit and XAUI backplanes?
My Answer is Absolutely NOT.
What should the Task Group be concerned with? In my opinion two things.
1) Developing an optimal signaling solution for those channels which
meet the criteria developed by the Channel Ad Hoc. These represent realistic
best practices for backplane construction. These channels do not
unnecessarily burden the potential solution with the power, cost and
complexity of legacy support.
2) Developing an Analytical Tool driven by S parameters, along the
lines of Stat-Eye which will allow the quantitative characterization of a
particular backplane as Complaint with 802.3ap 10 Gigabit Serial Signaling.
To aid the development of the market for 802.3ap 10 Gigabit Serial
Signaling, the most important thing is not burdening the standard with
legacy requirements. The most important thing is to provide a tool that will
allow systems designers to unequivocally know if an existing (or planned)
backplane will operate reliably at 10 gigabit serial rates.
Mike Lerer
Chief Architect Rapid Prototypes Inc.
Chairman Physical Link Layer Working Group of the Optical Internetworking
Forum
Chairman Hardware Working Group of the Network Processing Forum
Box 636 Londonderry, NH 03053
Cell: 603-548-3704