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

Re: [HSSG] 40G MAC Rate Discussion



Hi Ali,

I think we are in agreement with regards to the size/shape of a useful
physical interface for the server environment.  However, I would
suggest that you use the top surface area of the module as your figure
of merit when comparing one form factor to another.  Such that the X2
at 2779mm^2 versus the SFP+ at ~603mm^2 provides a factor of 460%.
This is why the X2 maximum power dissipation is rated at 4W and the
class 1 limit for SFP+ is 1W (again ~4x difference).  (Note that in
stacked module environments it becomes somewhat more difficult as the
ideal surface area comparison becomes chassis / air flow specific.)

"Imagine the technology required and time frame to support 100G on
servers requiring MMF/SMF?"

I have been imagining this extensively!  As stated earlier, using the
likely VCSEL array approach suggested by at least one of Jack Jewel's
presentations, the cost, size, and power dissipation of a 100G MMF
solution should scale sub-linearly other items being equal.  With
regards to the timeline, I'm encouraged by the increased performance
metrics from various 10GbE VCSEL suppliers which indicates that the
crosstalk penalties which Chris Cole alluded to for a parallel
interface should be conquerable if we are not overly onerous when
writing the standard.

With regards to an SMF interface I'm struggling to understand how this
would fit into a 40GbE server environment focused upon MMF.  I request
that any supporter of the 40GbE MAC rate provide additional insight as
to the (a) media, (b) order of magnitude volumes (eg. relative to PCI
express slots), (c) timeframe of the end user market.  I'd like to be
educated about the important metrics of this market.

In regards to an SMF interface for 100G, as mentioned in my 10km
presentation at the March plenary, one strategy for the "ultimate"
transceiver should be an integrated monolithic design for a
transmitter chip (sometimes referred to as a PIC) and receiver chip
each with four lanes running at ~25Gbit/s.  I am fearful that pushing
the small components industry into pursuing both a 4x10G PIC and a
4x25G PIC is impractical.  I can comfortably assert that the number of
companies equipped to design and enter into commercial production of
such a PIC is less than a handful.  I am leery of 40GbE due to a
potential pull for one or more SMF interfaces for the sake of a
"complete" standard.

I apologize for being long winded.

Thanks
--matt
(e) mtraverso@opnext.com

NOTE: This e-mail is being sent from my personal e-mail account rather
than my corporate e-mail address at Opnext due to default signature
files embedded in my Opnext e-mail account.

On 4/6/07, Ali Ghiasi <aghiasi@broadcom.com> wrote:
> Matt
>
> The current X40 is based on Xenpak MSA and was meeting the initial market
> requirement.  During the development of X40 at one point I proposed to
> start with X2.
> Currently LX4 module exist with each lane operating at 3.125 Gb/s, so
> the question here
> is what would it take to increase the laser speed to 10G/lane?
>
> Another data point we can use is the fact you can get LRM/LR SFP+ module
> with <1W,
> where X2 module is about 700% larger than SFP+ module.  Can you
> repackage 4 SFP+
> module in to an X2?
>
> Imagine the technology required and time frame to support 100G on
> servers requiring
> MMF/SMF?
>
> I did see Demo of QSFP at OFC operating at 5 Gb/s and some work has
> already been done
> to operate QSFP at 10Gb/s/lane.
> http://www.comsoc.org/livepubs/ci1/public/2007/mar/index.html
>
> Thanks,
> Ali
>
>
>
> Matt Traverso wrote:
> > Hi Marcus, Schelto,
> >
> > As my company is not a member of the X40 MSA, I should not comment on
> > this directly.  However, my limited understanding of the potential
> > market environment for a 40GbE physical interface suggests that the
> > X40 would not fit comfortably in the servers that Shimon is
> > envisioning.
> >
> > I believe that the server environment is more comfortable with a
> > physical interface which is similar in size and power dissipation
> > envelope to the QSFP, XFP, or SFP.  I believe that all of my
> > colleagues in the X40 MSA would agree that to achieve the power & size
> > for a SMF solution is challenging.  I don't believe that a lengthy
> > form factor discussion is appropriate or likely to be tolerated on
> > this thread, so this is why I'd like to understand as part of 40GbE's
> > BMP (broad market potential) effort the size / power / media
> > requirements.
> >
> > I'm going to stay out of the LAG discussion as my knowledge base
> > attenuates rapidly as we move up the protocol stack from the physical
> > interface.
> >
> > thanks,
> > --matt traverso
> > mtraverso@opnext.com
> >
> > NOTE: This e-mail is being sent from my personal e-mail account rather
> > than my corporate e-mail address at Opnext due to default signature
> > files embedded in my Opnext e-mail account.
> >
> > On 4/6/07, Marcus Duelk <duelk@alcatel-lucent.com> wrote:
> >>
> >> Hi,
> >>
> >> aren't there already 40G (i.e. 4x10G) transceiver out there ?
> >> I know at least of one transceiver company that is offering these
> >> devices today, there is also the X40 MSA group:
> >>
> >> http://www.x40msagroup.com/docs.html
> >>
> >> So I get the impression that the PMDs (including SMF) are
> >> already existing, it is maybe more a matter to have a MAC
> >> that supports this as one logical 40G pipe. But then we are
> >> back at that discussion what the difference to 4x10G LAG is,
> >> I assume ...
> >>
> >> Marcus
> >>
> >>
> >> Matt Traverso wrote:
> >> > I'd like to comment from an optical component / module vendor point of
> >> > view.
> >> >
> >> > Personally I'm not convinced that broad market potential has been
> >> > demonstrated, but... Operating under the assumption that the 40GbE
> >> > broad market potential is verified with end user input:
> >> > - As we heard/saw in Jack Jewel's presentation focused on the cost &
> >> > reliability of the MMF objective, extending from a 1x10G VCSEL to a
> >> > 10x10G VCSEL does not represent a linear cost increase -- similarly a
> >> > 4x10G would only be an incremental increase
> >> > - The dominant cost in a nx10G MMF interface is likely to be any
> >> > premium charged for the interface IC as well as costs associated with
> >> > the development quad laser drivers & quad amplifiers (or deka drivers
> >> > & amps)
> >> > - I'd like to hear a comment / perspective from the fiber
> >> > manufacturers on the utilization rate of the ribbon fiber strands.
> >> > For a 4x10G MMF approach presumably 8 strands in the 12 ribbon would
> >> > be used 4 for TX and 4 for RX.  For a 10x10G MMF approach it would be
> >> > 2 @12 with 10 @ Tx and 10 @ RX.  What does this do to the cost and
> >> > usage rate metrics of MMF cabling?
> >> > - Would an SMF PMD objective at 40GbE have broad market potential
> >> > (BMP)?  Here I am very skeptical
> >> > - Assuming that BMP was shown for an SMF PMD objective, I would
> >> > advocate a 2km serial 40Gbit/s scheme rather than a 4 lambda approach
> >> > as the transmission problems are not as severe
> >> > - This would represent the path that reuses the most technology and
> >> > allows for a compact & low power dissipation end solution
> >> > - As I have stated one of the primary impediments is the availability
> >> > of a low power interface IC -- this is the primary obstacle for OC768
> >> > (40G SONET/SDH) modules
> >> > - A 4 lambda x 10G at single mode would not simply be able to plug in
> >> > the work done on 802.3ae as the technical challenge of MUX/DMUX
> >> > optical loss and packaging would require a new round of investment
> >> >
> >> > In closing I'd like to see some supporting data for the Broad Market
> >> > Potential of 40GbE (including distance / media usage
> >> > comments/assumptions) that reflects the timeframe of standard
> >> > development -- eg. demand/need in 2009-2012.
> >> >
> >> > thanks,
> >> > --matt traverso
> >> > mtraverso@opnext.com
> >> >
> >> > NOTE: This e-mail is being sent from my personal e-mail account rather
> >> > than my corporate e-mail address at Opnext due to default signature
> >> > files embedded in my Opnext e-mail account.
> >>
> >> --
> >> ___________________________
> >> Marcus Duelk
> >> 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
> >>
> >>
> >
>
>
>