There are also a few options to consider
for the MM interface, each one using 1x technology (10G) at 850nm and assuming
a direct map from electrical lanes to optical lanes. These form what
I consider a reasonable collection of the lowest-cost alternatives that
result from combinations of SDM and WDM on modulo-12 cabling infrastructure.
I will list them sorted by the number of lanes.
10 lanes (64/66b encoded)
"10 SDM": (i.e. 1 wavelength) using 10 of 12 available
fibers in each direction, 24 fibers in total (4 spares, available for resiliency).
"5 SDM x 2 WDM": Two wavelengths (840 / 860) using 5 of
12 available fibers in each direction, 12 fibers in total (2 unused).
12 lanes (8b/10b encoded)
"12 SDM": (i.e. 1 wavelength) using 12 of 12 available fibers
in each direction, 24 fibers in total (0 spares).
"6 SDM x 2 WDM": Two wavelengths (840 / 860) using 6 of
12 available fibers in each direction, 12 fibers in total (0 spares).
"4 SDM x 3 WDM": Three wavelengths (835 / 850 / 865) using
4 of 12 available fibers in each direction, 12 fibers in total (4 unused).
The alternatives that use WDM each share
the ability to reduce the cabling costs by a factor of two in trade for
the added complexity in the transceiver. The historically greater
rate of decline in prices of transceivers compared to cabling leads to
the realization that the WDM alternatives provide a lower total cost of
ownership (TCO) in the time frame where the 100G market picks up. My
cost modelling shows that initially the pure SDM solutions will be most
cost effective for channel lengths up to about 50m. But with transceiver
price declines out-pacing cabling price declines, the WDM solutions become
more cost effective at progressively shorter lengths. In just a few
years time they offer lowest TCO. So while the pure SDM platforms
are the easiest path to market, they may not be the best long-term solution.
Some may find these statements odd coming
from a cabling systems provider, as the long-term lowest TCO solution cuts
our revenue per 100G channel in half. But it is my belief that developing
the true lowest-cost solution will help grow the market, hopefully sufficiently
to more than offset this fact.
CommScope Enterprise® Solutions
1300 East Lookout Drive
Richardson, TX 75082
Chris Cole <chris.cole@FINISAR.COM>
12/06/2006 08:28 PM
Please respond to
Chris Cole <chris.cole@FINISAR.COM>
Re: [HSSG] <HSSG-FO> Fiber Optic
Ad Hoc Teleconference 12/06/2006 Minutes
Good catch! For 10km reach
there will be no FEC. Below is an updated proposed study list for the SMF
From: Marcus Duelk [mailto:firstname.lastname@example.org]
Sent: Wednesday, December 06, 2006 5:36 PM
To: Chris Cole
Subject: Re: [HSSG] <HSSG-FO> Fiber Optic Ad Hoc Teleconference
thanks for the summary of today's call and the proposals. I have a question
regarding the last bullet item -- 2x56G DQPSK for single-channel intermediate
reach (2km w/o DCF, ~40km w/ DCF) at 1310nm.
Are you proposing 2x50G for a total of 100 Gb/s MAC rate, and therefore
the 2x56G was a typo ? Or is there some particular overhead (~12% ?) for
error correction ? I thought that this proposal was rather for a LAN PHY
(hence no FEC) !? If this is supposed to be with FEC then we are obviously
discussing a WAN PHY which should be rather at 1550nm than at 1310nm ...
Chris Cole wrote:
Based on today’s call, I would
like to suggest the following update to the proposed Alternatives Study
Call for email indicating attendance
Attendees: Dan Dove, Chris Cole, Steve Song, Greg McSorley, Larry Green,
Schelto Vandoorn,Milind Gokhale, Peter Dartnell, Xavier Clairardin, Winston
Way, Shashi Patel, Martin Carroll, Ralf-Peter Braun, Paul Kolesar, Itsuro
Chris Cole's Presentation
Moved through pg 1 and 2 readily, page three also discussed 25G(4 channel)
Peta r: Are you doing this in tandem with some work going on in the PCS,
or is this a module-side perspective?
Chris: System guys looking at MAC architectures, 1st gen MACs likely will
be 10G and 20GCDR is going to be pushing technology
Note: We need concensus on the PCS lane width as this may help direct optic
12x10G links (8G links with 8B10B coding) as an example versus 10x10.3
Petar is indicating that offering 12 channels would render parallel optic
solutions and WDM solutions with a common channel width
John Jaeger - Proposing x12 for WDM fiber?
Schelto/Petar - Latency is important to consider. 8B10B would provide lower
Chris: 4 Channel necessary?
Paul K: 4 channels have advantage for mux/demux being a binary value. But
acknowledges cost of increased data rate may not balance out.
Chris: 6/5/4 still open for 2X, 12/10 still open for 1X.
Chris: It would be nice to have 40Km solution that is leveragable from
10Km solution..we should think about it, but 40Km is currently not an official
Chris: 12x10 should be included in our consideration for multi-fiber solutions.
Chris: Sounds like we have agreement that nobody is calling for anything
other than 2 level signaling.
Petar: Is 12 channel much harder for WDM?
Chris: For DWDM, not. For CWDM.. Could be. The price of commonality on
the 12 channel between MM and SM is not readily understood as worth its
John J: Not interested in 12 channel WDM
Chris: Did not put up 10x10G @ 1310…does not fit in 1310 window.
Chris: Did not put 1310 in 1x technology because people are not really
doing that and would not be leveragable.
Chris: Sounds like we should add 12x10G on the electrical interface. We
should look at infiniband and see how we might leverage that.
Chris: Uncooled solutions provide substantial (40%) savings in cost.
Chris: We should add 2x56G DQPSK
Petar: Should we assume that PCS interface and PMDs will have different
Dan: Should we optimize the PCS interface for its ability to be leveraged
on MM fiber and then leave complexity to SM solution?
Chris: OIF offering to develop an electrical interface that can run up
Itsuro Morita'sPresentation =======================
DQPSK allows up to 50Km distance.
Petar: Why not measuring to 10e-12?
Itsuro: Did not use precoder, so expected error rate in receive could not
be measured. (not sure I captured this correctly)
Dan: RZ Carving?
Itsuro: For long distance transmission, RZ carving is additional modulation
stage driven by clock.
Chris: Is proposal for 10Km 1310 DQPSK?
Itsuro: Yes, if it can be applied to 1310, it would be good solution. Not
sure if its doable yet. Needs further study.
Chris: Don't believe DCF is practical.
Petar: If you use DCF, it would require tailoring of DCF to length… is
this the recommendation?
Itsuro: For 10Km, with fixed DCF may be possible. Needs study.
Chris: Are you going to study the 1310nm/10Km alternative?
Itsuro: Need to survey availability of technology to see if it can be done/shared.
Chris: Can you share which components are used?
Petar: Why 108 bit delay.
Petar: Modules? Should we try
to force them into the same form factor?
Dan: IEEE does not deal with implementations, but architectures.
Chris: Could we have 12 *and* 10 lane interfaces defined at MAC/PHY boundary?
Dan: To paraphrase; Can we have two PCSs and because they are highly digital,
embed them in the MAC/PCS chip and thereby provide a common differential
interface for either PCS with different lane counts? -- Seems possible
Larry: Meeting schedule for Monterrey?
Dan: Dunno. Please check the IEEE website. We will meet within the HSSG
and not as a separate breakout.
Dan: Please keep an open perspective on how to solve the MAC/PCS interface
solution when looking at PMD solutions with the desire that we come up
with something that works best for all cases.
If there any modifications
required to these minutes, please send me an email with specific recommendations
Monday Dec 18th at 10am
is our next meeting I will send out a notice with
specifics of the call soon.
Bell Labs / Alcatel-Lucent
Crawford Hill HOH R-237
791 Holmdel-Keyport Road
Holmdel, NJ 07733, USA
fon +1 (732) 888-7086
fax +1 (732) 888-7074