Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

RE: WWDM vs. 10Gb/s serial


I disagree with your suggestion that long haul 10xGbE is 6-10 years out.
6-10 months is more likely.

Already a "major" ISP has deployed several long haul 1xGbE CWDM systems up
to 1000 km in length using a combination of GbE transceivers and optics from
Pirelli.  In one installation from Quebec city to New York City it took 2
engineers in a Volkswagon bus 2 days to install this whole system.  On a
parallel set of fibers a more traditional SONET/DWDM system was installed by
the carrier.  It took them 3 months and a team of engineers to install that

The economics of long haul GbE is so compelling that I personally believe it
is this marketplace rather than the LAN that will drive 10xGbE development.
I  know of at least a dozen 1xGbE CWDN long haul systems that are currently
being deployed.  All of these users would quickly move to 10xGbE if it was

Several equipment suppliers are shipping CWDM 1xGbE systems that will go as
far as 400km

In long haul 10xGbE the cost comparison is with SONET, not LAN costs.


Bill St Arnaud
Director Network Projects



> -----Original Message-----
> From: owner-stds-802-3-hssg@xxxxxxxxxxxxxxxxxx
> [mailto:owner-stds-802-3-hssg@xxxxxxxxxxxxxxxxxx]On Behalf Of
> BRIAN_LEMOFF@xxxxxxxxxxxxxxxxxxxxxxxxxx
> Sent: Thursday, May 06, 1999 9:09 PM
> To: bgregory@xxxxxxxxx
> Cc: dolfi@xxxxxxxxxxxxxxxx;;
> stds-802-3-hssg@xxxxxxxx; dolfi@xxxxxxxxxx; twhitlow@xxxxxxxxx
> Subject: WWDM vs. 10Gb/s serial
>      I will try to respond to some of Bryan Gregory's remarks regarding
>      CWDM vs. 10-Gb serial. By the way, I will refer to it as WWDM
>      (SpectraLAN is HP's implementation of WWDM), since CWDM is
> apparently
>      used to refer to 400-GHz spaced telecom systems, and this has caused
>      some confusion among some people on this reflector.
>      First, let me say that I agree that long-term (say 6-10 years out) a
>      low-cost 10-Gb/s serial solution may be the simplest and lowest cost
>      solution.  That having been said, I think that with today's
> technology
>      (and for several years out) WWDM will be the lowest cost and most
>      useful technology for 10-GbE LAN applications.
>      Fiber:   A 4 x 2.5-Gb/s WWDM module in the 1300nm band should still
>      support useful distances of up to 300m on the installed base of 62.5
>      micron core fiber.  The SpectraLAN approach, like 1000LX, will
>      simultaneously support multimode and single mode applications (up to
>      10-km) with a single transceiver.  All 10-Gb/s serial
> approaches that
>      have been proposed (excluding multilevel logic) will require
> new fiber
>      to be installed in premises applications.
>      Laser Cost:  At 2.5-Gb/s, low-cost uncooled, unisolated DFB
> lasers can be
>      used with no side-mode suppression requirement (double moded
> lasers are
>      okay) up to 10km.  These lasers are readily available today
> in die form at
>      costs not that much higher than the FP lasers used in
> 1000LX.  Linewidth,
>      RIN, and Jitter requirements at 2.5-Gb/s are MUCH easier to
> realize with
>      high yield and low-cost electrical packaging than they are
> at 10-Gb/s (not
>      to mention 12.5 Gbaud). Optical isolation will probably be
> required to
>      achieve the necessary noise and linewidth requirements for a
> 10-km, 10-Gb/s
>      serial link (Lucent presented an unisolated FP solution for
> 1km.  The data
>      they showed for a 10km uncooled DFB link required
> isolation).  Given this,
>      I believe that the 4 lasers required for WWDM will be many
> times lower cost
>      than the single laser required for serial.
>      Optical Packaging Cost:  The 1000LX standard has forced
> transceiver vendors
>      to develop low-cost automated alignment and precision die
> attach systems
>      for aligning edge-emitting lasers to single-mode fiber. In our WWDM
>      solution, we are leveraging such a system to robotically
> assemble and align
>      our 4 lasers and MUX in a fast, low-cost process.  On the Rx
> side, only
>      multimode alignment tolerances are required to align the
> demux to the
>      detector array and glue it into place. The mux and demux
> optics themselves
>      are low-cost parts (many times lower cost than a
> micro-optical isolator).
>      The mux is a simple, unpolished, unpigtailed, silica waveguide chip
>      (several hundred devices on a standard 4" wafer).  The demux is an
>      injection-molded plastic optical part, requiring minimal
> assembly.  This
>      may sound complicated, but it is not expensive.  As we get
> further into the
>      standards discussions, we'll provide more details that
> should help convince
>      the skeptics that this is a realistic and low-cost solution.
>      Electronics:  WWDM at 2.5-Gb/s per channel works with
> existing low-cost Si
>      electronics.  10-Gb/s serial Tx and Rx IC's will require
> processes at least
>      4 times faster. Add to this the tighter jitter and noise
> requirements, the
>      poorer performance of dielectric circuit boards, the higher
> laser current
>      requirements (required to push relaxation oscillation
> frequencies 4 times
>      further out), and you have a difficult electrical problem to
> solve.  The
>      cost associated with the electronics and electrical
> packaging is likely to
>      be much higher than that for 4ch WWDM for several years.
>      Scalability:  Bryan made a good point that a 10-Gb/s serial solution
>      adopted now could be combined with WWDM later to provide even higher
>      capacity (e.g. 40 Gb/s).  Why not adopt the WWDM (4 x 2.5
> Gb/s) solution
>      now, when 10-Gb/s lasers and electronics are still very
> expensive, and then
>      in a few years, increase the channel rate to 10-Gb/s.
> Either solution for
>      10-GbE is scalable to 40-Gb/s when it is combined with the other.
>      Eye-safety:  The proposed power budget for SpectraLAN meets
> the Class I
>      eye-safety requirement by a comfortable margin.  At 1550nm
> it would be even
>      better, but increased fiber dispersion and the lack of
> well-characterized
>      fiber in the LAN make this a more difficult option.  It
> should be noted
>      that 4 lasers means 6-dB less eye-safe power available per
> laser, but at 4
>      times the speed, for a given IC process, a typical receiver
> will be less
>      sensitive by at least 6 dB, negating the eye-safety
> advantage inherent in
>      the serial approach.
>      "Inherent Simplicity":  A serial approach is "inherently
> simple".  The
>      question which we must answer over the coming year is which
> approach makes
>      the most practical sense from a performance and cost
> perspective, given the
>      technologies that are available today.
>      I hope I have at least provided a few reasons why 4x2.5-Gb/s
> WWDM might be
>      better than a 10 Gb/s serial approach, at least in the
> near-term.  There is
>      still a lot to be learned, a lot to be demonstrated, and an
> awful lot of
>      discussion to be had before one solution is chosen over another.
>      -Brian Lemoff
>      lemoff@xxxxxxxxxx
> ______________________________ Reply Separator
> _________________________________
> Subject: Re[2]: 1310nm vs. 1550nm -> Eye Safety + Attenuation
> Author:  Non-HP-bgregory (bgregory@xxxxxxxxx) at HP-PaloAlto,mimegw2
> Date:    5/6/99 9:32 AM
>      In response to Bill's email... regarding the EDFA issue, I'd imagine
>      that this would only be used in a small number of cases with
> a serial
>      10GbE approach.  I don't think it needs to be a core concern of the
>      group, but in some dark fiber trunking applications it can
> be useful.
>      I am most concerned about wavelengths vs. eye safety, and
> wavelengths
>      vs. fiber attenuation.  This could end up being a real killer.  Four
>      lasers @ 850nm or 1310nm put out quite a bit of light in an eye
>      sensitive range.  As I remember, four lasers at 1550nm offer a lot
>      more margin.  A single source at 1550nm could be very strong
> and still
>      meet the eye safe requirements.  This increase in power
> combined with
>      lower fiber attenuation would reduce some of the link distance
>      problems that we're bound to run into.
>      Also, long term I can't see how [4 lasers and an optical mux] + [4
>      photodiodes and an optical de-mux] would be better than a single
>      source and photodiode.  There is a lot of difficult
> packaging involved
>      in the CWDM approach.  I think the CWDM solution offers a
> quicker path
>      to market because most of that technology is available
> today. But long
>      term a single 10 Gb source (uncooled DFB without isolator) has a lot
>      of advantages.  It is intrinsically much simpler.  I think the board
>      layout and chip-sets will eventually support this as well.  If the
>      standard wanted to be able to scale beyond 10 gigs, even the serial
>      10Gb solution could allow further CWDM scaling.
>      Regards,
>      Bryan Gregory
>      bgregory@xxxxxxxxx
>      630/512-8520
> ______________________________ Reply Separator
> _________________________________
> Subject: RE: 1310nm vs. 1550nm window for 10GbE
> Author:  "Bill St. Arnaud" <> at INTERNET
> Date:    5/6/99 10:38 AM
> Hmmm.  I just assumed that 802.3 HSSG would be looking at 1550
> solutions as
> well as 1310 and 850
> I agree with you on longer haul links it makes a lot more sense
> to operate
> at 1550
> I am not a big fan of EFFA pumping.  It significantly raises the overall
> system cost. It only makes sense in very dense wave long haul systems
> typically deployed by carriers.
> CWDM with 10xGbE transcivers should be significantly cheaper.  That is
> another reason why I think there will be a big market for 10xGbE with all
> those transceivers every 30-80km on a CWDM system. However there is a
> tradeoff.  There is greater probablity of laser failure with many
> transceivers and the need for many spares.  I figure somewhere
> between 4-8
> wavelengths on a CWDM and transceivers is the breakpoint where it is
> probably more economical to go to DWDM with EDFA. Also EDFA is
> protocol and
> bit rate transparent.
> An EDFA will ..(edited).....  But EDFA window is very small, so
> wavelength
> spacing is very tight requiring expensive filters and very stable,
> temperature compensated lasers at each repeater site.  Also laser
> power has
> to be carefully maintained within 1 db otherwise you will get
> gain tilt in
> EDFAs. A loss of a signal laser can throw the whole system off,
> that is why
> you need SONET protection swicthing. But companies are developing
> feedback
> techniques to adjust power on remaining lasers to solve this problem.
> A single 10xGbE transceiver will .(edited)....??? Probably less.  So 6
> 10xGbE transceivers will equal one EDFA.  No problems with gain tilt.  If
> you lose one laser you only lose that channel, not the whole system.
> Protection switching not as critical, etc
> Bill
> -------------------------------------------
> Bill St Arnaud
> Director Network Projects