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

Re: [802.3BA] Discussion on 40G for => 10 km SMF



Hi Atsishi and others,
Let me comment on your points:
(1) COST - Today 40G serial technology is 70-100x the cost of 10G technology, so even with the assumption that 4x10G CWDM is 6x the cost of 10G, we still have that today's cost of serial 40G is >10x today's cost of 4x10G. 40G technology costs have been coming down lately, but not nearly as fast as the cost reduction of 10G. Matt has identified some improvements that can help, but not enough to bring the technologies to cost parity, and additional breakthroughs are needed before 40G serial could be considered a cost advantage to 4x10G. If one looks at the historical cost tends of 10G and 40G technology and asks when 40G will be cheaper than 4x10G, the answer would be "never" since the cost reductions of 10G technology have been so dramatic.
 
A lot of the cost will depend on volumes. If initial costs are low, volumes will be high and the costs will reduce quickly. If initial costs are high, as we expect for serial, volumes will be low and costs will not drop as quickly. In fact, a major concern should be that with serial, by the time the interfaces are cost effective enough to start selling in volume, the market opportunity for 40G will have passed and the market will have migrated to 100G (using 4xnG WDM, the parallel interface types that serial advocates were hoping to avoid) and there will not be much interest to go back and deploy 40G. Allesandro remarked that a choice would be to delete the 40 GbE SMF objective, and while I don't want to do that either, I also don't want to meet the objective in a way that is unlikely to receive wide market adoption due to high cost.
 
(2) TIME - there is no question that both serial and CWDM can be built if cost is not a factor. But if one simply wants to build a 1310nm version of an OC-768 type interface, it will be way too expensive for the Ethernet market and will not see significant adoption. Significant work, including some (as yet unidentified) breakthroughs are needed to make a 40G serial implementation that would be attractive in cost to the Ethernet market.
 
(3) Power Consumption - Today, 40G SerDes are very power hungry. It will take evolution in CMOS processes to make 40G Serial lower power than 4x10G CWDM.
 
(4) Electrical Interface (I won't comment on the size) - The electrical interface (host board to module) for 4x10G CWDM and Serial would be exactly the same. There is no feasible 40G electrical signaling that can be used for this, so the electrical interface is 4x10G in any case, which could be used to drive the individual optical lanes for CWDM. For serial, these 4 electrical lanes will need to be bit-multiplexed on the module (functionally a 4:1 PMA).
 
(5) Risk for wavelength failure - I would love to see a FIT analysis of this, looking at the probability of failures of the various optical and electronic components of the interface. Chris has asserted that the reliability of the optics is extremely high. Even if the optics are serial, much of the electronics are parallel, and even the optics is logically parallel since it consists of four bit-muxed VLs. There is no redundancy in either case - if you lose any part of the optics, any electrical lane, lose your lane marking logic, the interface is completely dead (you cannot reconstruct a signal from 3 of 4 optical lanes - but neither can you reconstruct a signal from a serial interface that has a stuck bit on a VL or electrical lane in the host board to module interface). So what would be interesting to know is if there is a significant difference in the total interface reliability based on the choice of 4x10G CWDM or serial.
 
(6) Speed breakthroughs - I don't think is correct that we have seen regular speed breakthroughs. Things moved quickly up through 1GbE. 10GbE started out very high in cost, and volumes took far longer to ramp than with previous generations. In transport networks, we were increasing speed by a factor of four roughly every four years (OC-3/STM-1, OC-12/STM-4, OC-48/STM-16, OC-192/STM-64, OC-768/STM-256). In both cases, we seem to see slowing as we reach the limits of what can be done cost effectively in a serial optical interface. 40G technology was introduced comercially in the market 6-7 years ago, and adoption has yet to reach the pace of 10G technology due to relatively higher cost. Nobody is discussing introduction of 160G technology. Some of the breakthroughs to make 40G more cost effective involve the use of advanced modulation formats to reduce the bit-rate, e.g., DP QPSK encodes data on two phases of light per polarization at one quarter the bit-rate. But this is effectively a four-lane parallel interface, the same as we are trying to do for 40 GbE! For the Ethernet market, if one wants to do a four-lane parallel interface, it is more reasonable to do this with simple NRZ coding on parallel wavelengths instead of the type of sophisticated transceiver one would use for amplified multi-wavelength systems going thousands of km where you want to do sophisticated things to put all of what are actually parallel lanes onto the same wavelength. As the limits of a technology are reached, additional capacity is achieved with parallel interfaces.  Don't forget that the most successful interface that 802.3 has ever developed (the 1 GbE copper interface) is a parallel interface because it is not feasible to cost effectively transmit 1 Gb/s serially over an electrical interface. Nobody thinks of this interface as comprised of parallel lanes because it is just one RJ-45 connector at the end of a cable. Nobody worries that individual lanes might fail because the interface as a whole is quite reliable.
Regards,
Steve
 


From: Atsushi Takai [mailto:atsushi.takai@xxxxxxxxxx]
Sent: Thursday, July 31, 2008 1:18 AM
To: STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3BA] Discussion on 40G for ="" 10 km SMF

John
 
Thanks.
I may understand.
If you will find any violation below, please let me know.
 
 
TF members
 
I am supporting 40G serial.
My view on 40G Serial and CWDM is as below.
 
(1) COST
 
We can achieve less than 4x10G cost using 40G serial
while we need some breakthrough technology using 40G CWDM
 
I agree 40G CWDM will be 4-8 times of 10G as written in "cole_04_0708" page 8.
And most likely 6 times.
Using current technology, it is difficult to achieve less than 4 because we have to pay for wavelength control.
In case of 40G serial, module structure is the same as X2.
I believe we can achieve less than 4 times cost for each part in the 40G module comparing 10G serial.
We are waiting 3rd generation SERDES and 2nd generation of driver and TIA IC to achieve low cost 40GbE serial.
I am sure that industry is working on these devices.
 
(2) Time
 
There exists plural 40G serial module vendors today using 1550-nm EA-DFB.
1310-nm EA-DFB is easier because we can neglect dispersion problem.
Thus 40GbE serial is possible.
And cost reduction plan is visible.
I also sure 40G CWDM will be available if market will accept larger than 4 times cost.
Cost reduction plan will follow the same as 10G and plan for less than 4 times is invisible today.
 
(3) Power consumption
 
We need also breakthrough to achieve less than 4 time power consumption using CWDM.
Power consumption reduction plan is invisible
We are sure we can achieve less than 4 time power consumption using serial in future.
 
(4) Size
 
I am not sure it is too early to talk 40G electrical interface.
But 40GbE serial has possibility for XFP or SFP+ size.
I did not hear LX4 XFPor SFP+ due to pin constraint and power consumption.
 
(5) Risk for wavelength failure
 
WDM technology has always the risk for wavelength failure, while serial does not have.
We may resolve some way for 40G CWDM because of high reliability of long wavelength optical devices.
 
(6) Another aspect
 
We made speed breakthrough every 4 or 5 years in the past.
AND
We started to deliver 10G modules in 1997.
The 300-pin MSA started in 2000.
10GbE was issued in 2002.
40G is the next milestone for technology evolution and now is a little bit behind the past trend.
(We had unhappy period that every progress seemed to stop.)
Today 40G is the technology to challenge and overcome for both optical and electrical technology.
I  think IEEE should not avoid such technology evolution.
 
I believe IEEE should take 40GbE serial.
 
=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
Atsushi Takai
Marketing Division, Opnext Japan, Inc
 
=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
----- Original Message -----
Sent: Wednesday, July 30, 2008 8:30 PM
Subject: RE: [802.3BA] Discussion on 40G for ="" 10 km SMF

Atsushi,

First, let me say that the email below is intended to make sure that this group does not stray in a direction that would ultimately lead to the violation of IEEE-SA Antitrust and Competition Policy.  As chair, I am remaining neutral on the TF’s technical decisions.

 

Regardless of the decision that this body makes, the market may demand that both solutions are developed anyway.  

 

Please note in the following document from the IEEE, Promoting Competition and Innovation:

What You Need to Know about the IEEE Standards Association’s Antitrust and Competition Policy,” which may be found at http://standards.ieee.org/resources/antitrust-guidelines.pdf.   Please note the following statement:

 

For example, selecting one technology for inclusion in a standard is lawful, but an agreement to prohibit standards

participants (or implementers) from implementing a competing standard or rival technology would be unlawful – although as a practical matter, a successful standard may lawfully achieve this result through the workings of the market.”

 

As I have phrased it to the Task Force, the Task Force makes decisions about what it is going to do, it does not make decisions about what it is not going to do.

 

Regardless of the decision that this Task Force makes, it is very easy to envision both implementations getting developed in the industry.  Given the need stated by CWDM supporters for a near term solution, it is easy to envision an industry effort happening if the TF goes serial.  It is just as easy to envision a new CFI happening for a serial solution if the TF chooses to go CWDM.  

 

Also, as a point of clarification, as I am currently looking at the presentation for another discussion, you may wish to refer to Flatman_01_0108 (http://grouper.ieee.org/groups/802/3/ba/public/jan08/flatman_01_0108.pdf), which is a survey of data centers that Alan Flatman did that shows 40G being deployed in access-to-distribution links in 2010.

 

Regards,

 

John

 

 

 

 


From: Atsushi Takai [mailto:atsushi.takai@xxxxxxxxxx]
Sent: Wednesday, July 30, 2008 2:26 AM
To: STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3BA] Discussion on 40G for ="" 10 km SMF

 

Hi Mark

 

I agree that DC application is cost sensitive.

As many people agreed in some presentations at meeting, CWDM will be cheaper at near term and serial will became cheaper in maybe 2011 or beyond.

That means transceiver supplier have to develop CWDM in 2009 timeframe and serial in 2010 or 11.

This development will cost much. And I do not think CWDM cost in 2009 or 2010 will be cheaper than 4x10G.

Even more DC application users can choose 8x10G CWDM that has more bandwidth.

I do not think CWDM has superior merit for DC application.

Also I think there was a presentation that said that DC will start install 40G in 2015 or beyond.

 

=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
Atsushi Takai
Marketing Division, Opnext Japan, Inc

 

=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=

----- Original Message -----

Sent: Wednesday, July 30, 2008 4:08 AM

Subject: Re: [802.3BA] Discussion on 40G for ="" 10 km SMF

 

John,

 

I agree some reflector discussion would be helpful on this topic.

 

I characterize the discussion I heard around the choice of PMD as really boiling down to a debate on the primary applications and the drivers for those applications.

 

The original justification for adding the 40G SMF objective was primarily based on the application of Data Center inter-switch links.  There was also other applications such as for use in interconnecting to OTN equipment which is good for BMP.  In both cases lower cost solutions are preferential.  The debate appears to revolve around what other assumptions there are around market timing, technology risks, cost projections and operational issues etc.

 

To me the simple view is that to achieve low cost, you need higher volume.  Higher volume is achieved by the having a solution that addresses the largest primary application and as many others as possible.  I am assuming that the primary application is still the DC.

 

Since DC applications are inherently much more sensitive to cost, a near term low cost solution is needed or else the application will likely not be adopted.  In this case, if 40G SMF PMD is uneconomical in the near term then the DC users will likely stay with nx10G as long as possible and then presumably assess the 40G/100G economics at some later date.

 

The argument for adopting serial technology now is that the potential higher volume of the DC application will trigger the necessary development investments now and drive the cost of that technology down so we will ultimately get it to the low cost solutions needed.  My concern is that the timing and cost windows needed for the DC application do no fit with that model and we would end up with little adoption in that market and end up with a lower volume, higher cost PMD which is what we would all like to avoid.

 

Mark

 

 

 


From: John DAmbrosia [mailto:jdambrosia@xxxxxxxxxxxxxxx]
Sent: Monday, July 28, 2008 6:53 PM
To: STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
Subject: [802.3BA] Discussion on 40G for ="" 10 km SMF

Dear Task Force Members,

Per Motion #9 from July, the editorial team is working on creating a “a draft based on adopted baseline proposals for circulation prior to the September 2008 interim meeting

.”  Unfortunately, at the July meeting the Task Force did not reach consensus on a baseline proposal to satisfy the 40G over => 10km SMF objective.    Therefore, in September we need to reach closure on this issue.  

 

With that said, I would like to strongly recommend that the TF make use of the reflector to discuss the various issues of debate that have been going on, both during the meetings and during offline discussions. 

 

Let’s use the next several weeks to have meaningful debate so we can reach consensus at the September meeting.

 

Regards,

 

John D’Ambrosia

Chair, IEEE P802.3ba Task Force