RE: PMD discussion
Ed, Bruce,
I think we all ought to concern ourselves with
the needs of the end customer and system integrator.
In this regard, I believe Bruce has been doing some
serious listening and has some compelling arguments.
While agreeing with the assessment of the customer
needs which Bruce has stated, I also see many vendors
of optical transceivers approaching this issue with
a certain trepidation.
It may be an 850nm serial or parallel transceiver is
the lowest cost implementation for very short haul 10G
links. It is apparent the goals of the 10GbE standards
effort do not extend to this short reach style connection.
What Bruce is telling us is there may not be a place in the
standard for all of the PMD variants which primarily address
the short control room type applications. He is also leaving
open the possibility of using 850nm serial and/or parallel
connections for switch to switch and other connections up to
100m.
We might consider as a group of committed 850nm devotees if
our activities at promoting 850nm technology aren't going a
bit too far here, since we do not meet the group's stated
objectives for ditance and installed fiber base. If we do not
consider this option, must ask the IEEE 802.3 to reexamine their
goals and allow for the limitations of the hardware we are
currently proposing as a solution.
I invite your responses/comments.
Regards,
Pat Gilliland
patgil@xxxxxxxxxxx
--------------------------------------------------------------
At 02:58 PM 5/30/00 -0700, you wrote:
>
>Ed:
>
>As far as I am concerned comparing 850 nm WDM with 1310 nm WDM is like
>comparing apples and oranges because they no not address the same customer
>segment.
>
>My customers want to run 10 GbE in the following spaces:
>
>1) Very, very short reach for connections between switches in the same
>room. For this application,parallel solutions seem to work fine
>
>2) Building backbone on installed 62.5 micron MM fiber. For this
>application, only one proposal works: 1310 nm WDM
>There is no threshold of pain in regard to cost, since if I cannot sell to
>my installed base THIS BUSINESS IS DEAD ON ARRIVAL.
>
>The 850 nm WDM PMD could cost 1/10 that of the 1310 nm part but that does
>me no good because the market is much smaller.
>
>Even if there were a substantial cost difference, because I expect cost to
>be a function of the experience curve, and the costs of the 1300 nm WDM to
>come down as volume ramps.
>
>And volume should ramp more quickly for a part that supports the installed
>base of 62..5 MM fiber. Remember, customers resist FORKLIFT UPGRADES. If
>you examine all the demonstrations to date of 10 gbE technology at trade
>shows, they are all demonstrations of upgrades to technology and products
>customers already have installed. This is no accident. We must support the
>installed base to have this market take off quickly.
>
>And by the way, if I am going to recommend new fiber for future proofing, I
>might as well recommend SM fiber.
>
>3) Medium long reach ( 10 to 40 km). Several PMDs cover portions of this
>spectrum. 1300 nm serial seems most optimal.
>
>4) Very long reach (over 50 km). 1550 nm serial seems the most optimal.
>
>Thanks for listening
>
>Bruce
>
>Bruce Tolley
>Enterprise Line of Business
>Cisco Systems
>
>At 03:25 PM 5/30/00 -0400, Cornejo, Edward (Edward) wrote:
>
>>Jack, et al,
>>
>>I could almost live with the list you have below, however, I thought Mr.
>>Tolley mentioned the importance of the embedded base(62.5um) up to 300m. If
>>this is the case, I would change the WDM from the 850 to 1310 window. The
>>only issue is cost between the two WDM solutions and I am sure there is a
>>difference, but by how much?
>>
>>Also, systems folks need to tell us what their threshold of pain is for
>>delta differences in cost. Is it 10%, 15%, 20% or 25%(0% is not an option,
>>sorry). This could make our decision easier.
>>
>>Regards,
>>Ed
>>
>>
>>
>> > ----------
>> > From: Jack Jewell[SMTP:jljewell@xxxxxxxxxxxxx]
>> > Sent: Tuesday, May 30, 2000 2:59 PM
>> > To: '802.3ae'
>> > Subject: RE: PMD discussion
>> >
>> >
>> > Walt (and others who have responded in the interim),
>> >
>> > Although I am encouraged that the group will come to some agreement
>> > without
>> > delaying our schedule, there is still some way to go. Given the breadth
>> > of
>> > the 5 objectives set forth, it seems almost certain that a 3-PMD set will
>> > leave at least one objective being greatly under-optimized. I would hate
>> > for that to be the highest volume product. Following the theme of
>> > optimizing for each space presented Thursday by Steve Haddock (and
>> > departing
>> > from his presentation of only 3 spaces), I would see the following as the
>> > optimal solutions for each Objective. Given the concensus that the 2km
>> > and
>> > 10km objectives are best served by a common PMD, we are really discussing
>> > 4
>> > distinct Objectives. Most of the debate is focused on the
>> > shorter-distance
>> > objectives, so this note focuses on them. I recommend the 4-PMD set
>> > below.
>> >
>> > SMF up to 40km - 1550nm Serial
>> >
>> > SMF up to 10km (incl 2km) - 1310nm Serial
>> >
>> > MMF up to 300m - 850nm Serial
>> >
>> > Installed MMF up to 100m - 850 CWDM
>> >
>> > Highest Volume Product Space
>> > The highest-volume product space here is almost certainly the MMF up to
>> > 300m. This is due to most of the transceivers being put into new
products
>> > with new fibers. An interesting comment was made last Thursday regarding
>> > 100Mb Ethernet in which 3 PMDs were spec'd in order to accommodate TX
>> > grade
>> > cable as well as the installed base of T2 and T4 cable. Subsequently the
>> > T2
>> > and T4 were more or less abandoned and virtually the entire market went
>> > for
>> > the TX-based PMD. [I did not participate in that process; this is my
>> > interpretation of the comments made last Thursday. If this is
inaccurate,
>> > I
>> > apologize.]
>> >
>> > Cost "Survey"
>> > There have been several comments today regarding costs of different PMDs.
>> > The cost comparisons which I (and Paul K) presented were NOT the
result of
>> > an "unscientific survey." They are the average of estimates which were
>> > circulated on the reflector. No one was excluded from presenting their
>> > own
>> > numbers. I believe a great deal of thought was put into each estimate.
>> > Are
>> > the numbers accurate? Of course not. Are they all consistent enough to
>> > represent the general picture (which is how they were presented)?
>> > Absolutely. They have been presented publicly and privately with no
>> > voiced
>> > disagreement.
>> >
>> > Relative Costs of Serial and WDM
>> > It has been suggested and sometimes even stated outright that a WDM
>> > solution
>> > will be as low-cost or even lower-cost than a serial solution. This is
>> > unsupportable. I put this question to (and look for responses from)
>> > vendors
>> > who have manufactured and sold transceivers in any reasonable volume.
>> > (These are the ones who know transceiver costs better than anyone.) Does
>> > your optical subassembly cost more than your ICs? In expectation of a
>> > unanimous "Yes" to that question, comes the next question. In order to
>> > move
>> > to higher performance levels while minimizing the cost increase should
>> > you:
>> > 1) increase the complexity of your optics; or 2) increase the complexity
>> > of
>> > your ICs? This is why we aren't seriously discussing a WDM-only set of
>> > PMDs. WDM is a good way to get more data over a fiber than otherwise
>> > possible - but it's used only when simpler approaches cannot be used. As
>> > for the ICs, there were a lot of IC vendors at the meeting having
>> > aggressive
>> > goals and impressive capabilities. Assuming a "market entry" for 10GbE
>> > products at the end of this year, it is reasonable to forecast that the
>> > cost
>> > to produce a serial product will be less than its equivalent WDM product
>> > at
>> > the time of market entry or within 6 months after that. I.e. for
>> > essentially the entire product lifetime.
>> >
>> > The 4 PMDs recommended above are not biased in order to favor my
company's
>> > choice of PMDs; rather my company's choice of PMDs is based on a hard
view
>> > of the markets and technologies required to address them in an economical
>> > fashion.
>> >
>> > Jack
>> >
>> >
>> >
>> > -----Original Message-----
>> > From: Walter Thirion [mailto:wthirion@xxxxxxxxxxxx]
>> > Sent: Sunday, May 28, 2000 11:58 AM
>> > To: '802.3ae'
>> > Subject: PMD discussion
>> >
>> >
>> >
>> > First of all, thanks to everybody that presented PMD proposals at the
last
>> > meeting. I've sent my presentation to David Law, so it should be
available
>> > on the web site in the next couple of days.
>> >
>> > In listening to the discussion after my presentation and then going
around
>> > and talking to people, it feels to me like we're starting to converge.
Not
>> > there, yet, but making progress.
>> >
>> > The equipment manufacturers made it pretty clear they would like to
see no
>> > more than 3 PMDs in the standard. The PMD vendors have some concern that
>> > using only 3 PMDs may sub-optimize certain objectives, however, they
could
>> > support the 3 PMD position if it is made clear which 3 PMDs the equipment
>> > oems want.
>> >
>> > Based on an informal straw poll and anecdotal evidence, my opinion is the
>> > first choice would be the set:
>> > ________________
>> > 850 nm WWDM
>> > 1310 nm WWDM
>> > 1550 nm Serial
>> > ________________
>> >
>> > If that set isn't feasible, then the 2nd most popular choice is:
>> > ________________
>> > 850 nm WWDM
>> > 1310 nm Serial
>> > 1550 nm Serial
>> > ________________
>> >
>> > Thoughts, feedback?
>> >
>> > Walt
>> > ___________________
>> > Walter Thirion
>> > Chair, IEEE 802.3ae PMD Sub-Task Force
>> > 301 Congress Ave.
>> > Suite 2050
>> > Austin, Texas 78701
>> > Voice: 512-236-6951
>> > Fax: 512-236-6959
>> > wthirion@xxxxxxxxxxxx
>> > ___________________
>> >
>