Re: [802.3_NGEPON] ONU ASIC and wavelength plans

[Barry Colella <barry.colella@xxxxxxxxxxxxxxxxxxx>]

I agree that the issue is to decide whether to have 25G only ONU be able to operate on more than one wavelength (and 50G on groups of two). 

 

This will then help to drive the next decisions regarding upstream wavelength selections.  We should keep in mind that Multiple wavelength ONU and tunability may be regarded as undesirable from a cost and OPEX perspective but there is no reason for the standard to specifically disallow (or make it unfeasible) these features.  One could simply purchase all the same color for the their 25G ONU and vote with their pocket book by driving pricing for that selected wavelength.  I understand that this would cause some pain on the schedulers in the MAC devices but the enhanced coverage provided by many low cost ONU on the same PON port  has value, especially when looking to drive volume.

 

NGPON2 suffers in that it required tunability from the outset as an integral part of the offering, thereby introducing a discontinuity in the evolutionary cost curves of access optics and consequently slowing down deployment.

 

A possible example scenario could be:

 

A base 25G ONU could start out as a fixed wavelength at lambda 0

Next step would be a tunable 25G ONU that adds 30% optics cost but could work on any of the 4 channels (or 3 remaining channels if we decide lambda 1 is fixed CWDM)

This would facilitate network expansion at the rate of demand

The 50G ONU could follow accordingly

 

-Barry

 

From: Marek Hajduczenia [mailto:marek.hajduczenia@xxxxxxxxx]
Sent: Friday, February 19, 2016 6:19 PM
To: STDS-802-3-NGEPON@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_NGEPON] ONU ASIC and wavelength plans

 

Shawn,

 

If we assume that 25G devices can use any of ls defined in the system, then the number of combinations of devices that can be present on the PON grows very quickly. It is then also natural to assume that 50G devices have the same flexibility, and this is what we would end up with:

 

-          4 different flavors of 25G ONUs

-          At least two different flavors of 50G ONUs (l0 + l2 and l1 + l4), but then nothing prevents other combinations as well (l1 + l2, as an example )

-          One flavor of 100G ONUs (all ls supported)

 

While it might look very attractive from specification perspective (pretty much anything goes), it becomes unmanageable from operational perspective very quickly ? bandwidth on each wavelength would have to be managed separately, leading quickly to situations where higher data rate ONUs might be starved for bandwidth because of presence of lower data rate ONUs using the same wavelength. Furthermore, this also increases OLT scheduling complexity, where each wavelength will have devices operating at different speeds content for access.

 

Also, I think we likely read Glen’s bullet points you showed below slightly different. An operator would not start deploying 50G devices until it was necessary to do so, and then device would not be operated at reduced wavelength count unless there was an operational need to do so (failure, maintenance, etc.). Deploying a 50G devices and then cutting their capability to 25G is like buying a racing car and then driving it only on the driveway ? yes, one can certainly do that, but it does not make much sense. As mentioned on the call, I believe operators will deploy in the future 25G systems until 50G deployments are driven by customer demand (whether it is poster speed or other conditions, it is beside the point). Power consumption argument is a hard sell for me ? the ONU power consumption is next to nothing compared to WiFi enabled home gateways blasting high speed WiFi on multiple antennas 24/7, irrespective of whether there is data to transmit or not. In business environment, it is even less of an argument, given that an ONU is typically coupled with an edge device that can easily consume 50 times more power or so.

 

As you can probably tell, I am all for simplicity and as few options as possible. This will lead to more consolidated market, fewer transceiver options, and hopefully ? lower cost of end devices.

 

Now, as far as I straw polls you suggest are concerned, I must admit that I do not know how to respond to some of them. The data rates listed in brackets are quite arbitrary and are applicable to one specific deployment scenario. Rather than tie in decisions to customer data rates, let’s just decide whether:

 

-          25G-EPON operates on only one of available wavelengths (l-0-downstream, l-0-upstream)

-          50G-EPON operates on only one pair of available wavelengths (l-0-downstream, l-2-downstream, l-0-upstream, l-2-upstream)

 

Obviously, 100G-EPON would then operate on all downstream and upstream wavelengths, no decision needed there. For 50G-EPON, we can use wavelength number 2 or 1, I do not have any preference here.

 

Marek

 

From: Shawn Esser [mailto:shawn.esser@xxxxxxxxxxx]
Sent: Friday, February 19, 2016 5:45 PM
To: STDS-802-3-NGEPON@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_NGEPON] ONU ASIC and wavelength plans

 

Marek,

 

Thanks for the feedback.

Note that I understand Jorge’s requirement that 50G ONU’s and 100G ONU’s will be needed for speed test and actual customers with higher data rate services than what 25G ASIC can deliver. My comments are only pertinent when network capacity > 25G is needed on the PON Group, but the peak data rate services can still be accommodated by a 25G ASIC -> in Jorge’s example, I believe this would be max Flagship speed of 8Gb and max Billboard speed of 12Gb. Maybe this scenario would not exist so then the rest of this discussion is moot.

 

If not decided yet, we need to decide if the 25G ONU (25G ASIC) is only available at λ0, and 50G ONU (50G ASIC) is only available at λ0 and λ1. If this is the case, then tunable optics in the ONU does not provide any value and the standard/product requirements would be simplified. If the 25G ONU (25G ASIC) is also to be available on λ1, λ2, and λ3, then tunable optics would provide value to minimize part numbers.

 

Marek, I believe you are suggesting to deploy 50G ONU and 100G ONU’s when more capacity is needed on the PON Group, even if the peak service rate to all the consumers only needs to be max 8G/12G (per Jorge example). The 50G ONU and 100G ONU’s may only operate on 1 wavelength pair in this case so the optics in the these ONUs for the other wavelengths are not utilized. I think it goes to this part of Glen’s presentation:

 

 

In this case, (#1) 50G ONU operating at 1 wavelength may be more expensive than a 25G ONU with tunable optics. Certainly (#2) 100G ONU operating at 1 wavelength will be more expensive than a 25G ONU with tunable optics.

 

Maybe these should be some straw poll questions:

・         Does the standard need to accommodate 50G network capacity in a PON Group with peak data rates that can be served by 25G ASIC in ONU (max 8G/12G to consumer)?

・         Does the standard need to accommodate 100G network capacity in a PON Group with peak data rates that can be served by 25G ASIC in ONU (max 8G/12G to consumer)?

・         Does the standard need to accommodate 100G network capacity in a PON Group with peak data rates that can be served by 50G ASIC in ONU (max 16G/25G to consumer)?

・         Shall the 25G ONU (25G MAC) be only available at wavelength pair λ0?

・         Shall the 50G ONU (50G MAC) be only available at wavelength pairs λ0 and λ1?

 

Thanks,

Shawn

 

From: Marek Hajduczenia [mailto:marek.hajduczenia@xxxxxxxxx]
Sent: Friday, February 19, 2016 8:20 AM
To: Shawn Esser; STDS-802-3-NGEPON@xxxxxxxxxxxxxxxxx
Subject: RE: [802.3_NGEPON] 802.3ca (100G-EPON) bi-weekly consensus - call for agenda items

 

Thank you, Shawn,

 

I think we are in agreement on the 25G assumptions, i.e., short time to market, single wavelength system, with fixed wavelength. Regarding your question: “Does standard need to specify fixed/tunable or leave it to the market to determine?” ? the answer would be yes, we do need to decide. 802.3 standards are developed to guarantee interoperability between different implementations, so leaving things such as wavelength allocation for “market to determine” is not going to work out. There would be no guarantee that ONU from vendor X would then work with OLT from vendor Y without additional interoperability testing at the operator lab, which is what we want to avoid.

 

Regarding the second bullet on page 2, I am not sure that I agree with the statement: “which may justify costs of tunable optics”. If I recall correctly, the statement on the call was in the lines of being able to accept higher cost devices on generation 2 and 3 only because the migration to higher speeds will be drive again by top tier customers, where it is easier to accommodate the higher cost of CPEs. This does not mean, however, that we need extra capabilities just for the sake of having them. Very much rather than dabble with tunability, I would prefer to have the selectable wavelength enable/disable capability that Glen presented yesterday. This feature has functional and operational justification, versus added management and operational complexity of managing wavelength allocation to individual customers. Personally, I would very much prefer to maintain current speed progression:

-          Deploy customers on 25G.

-          When we start running out of steam on 25G, we move selectively high bandwidth customers to 50G, freeing bandwidth on 25G for more connections.

-          Rinse, wash & repeat for 50G and 100G in the future.

This progression can be easily converted into an operational process, taking away unnecessary flexibility which will end up not being used.

 

As far as slide 3 is concerned, I do not see much value in this arrangement. It is not a true 100G system in that a single customer will always be limited to 50G. Operationally, it would be either a nightmare (if fixed colored optics is used) or more expensive / operationally more complex (when tunable optics is used).

 

Marek

 

From: Shawn Esser [mailto:shawn.esser@xxxxxxxxxxx]
Sent: Thursday, February 18, 2016 6:01 PM
To: STDS-802-3-NGEPON@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_NGEPON] 802.3ca (100G-EPON) bi-weekly consensus - call for agenda items

 

Re-sending with copyright language removed.

 

Thanks,

Shawn

 

From: Shawn Esser
Sent: Thursday, February 18, 2016 5:21 PM
To: STDS-802-3-NGEPON@xxxxxxxxxxxxxxxxx
Subject: RE: [802.3_NGEPON] 802.3ca (100G-EPON) bi-weekly consensus - call for agenda items

 

Attached are my notes, comments and questions from today’s review of Glen’s presentation. Great job Glen in spurring great discussion! Hopefully this will drive us to making some decisions soon.

 

Thanks,

Shawn

 

From: Dai, Eugene (CCI-Atlanta) [mailto:Eugene.Dai@xxxxxxx]
Sent: Wednesday, February 17, 2016 12:26 PM
To: STDS-802-3-NGEPON@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_NGEPON] 802.3ca (100G-EPON) bi-weekly consensus - call for agenda items

 

The tunable part of NG-PON2 has been often used to compare with 100G EPON. It is true that both NG-PON2 and 100G EPON are build on WDM. However, there is a fine difference on how the WDM is used.

 

NG-PON2 is build on multiple single-channel TDM, and the underline WDM is a wavelength-select WDM PON. Using tunable optics is an architecture to build wavelength-select WDM PON. In the architecture each ONU only needs ONE tunable laser to tune a given wavelength.

 

100G EPON is build on a parallel architecture. Like in the 100G and 400G Ethernet, WDM is used in a broadcast way. In the PON world, one can say that the 100G EPON is build on a wavelength-broadcast WDM PON. If tunable opticas are used in this architecture, an ONU may have multiple tunable optical to tune to multiple channels. Therefore, the cost will be (much) higher than that in NG-PON2.

 

Eugene  

 

---

From: Francois Menard <fmenard@xxxxxxxxxxx>
Sent: Wednesday, February 17, 2016 9:53 AM
To: STDS-802-3-NGEPON@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_NGEPON] 802.3ca (100G-EPON) bi-weekly consensus - call for agenda items

 

Marek,

 

While much has been published on the topic of InP monolithic integration and Datacom-grade Hybrid Integration on Silicon much less has been on Telecom grade hybrid integration on Silicon. There InP yield will remain that of single channel per InP die (high) and the issue of  costs will be solved by way of packaging cost efficiently.

 

NG-PON2 requires tunable receivers and I am convinced we will see a lower cost solution there fist.

 

An NG-EPON ONU with a single channel will need to block 3 out of 4 channels. The question will be 'which ones' and 'why must this be static' ?

 

And my point remains for the case of more than 4 channels.

 

F.

 

Sent from my mobile device. Apologies for any typo.

On Feb 17, 2016, at 09:07, Marek Hajduczenia <marek.hajduczenia@xxxxxxxxx> wrote:

Francois

 

What is not clear to me from your email is why we would need to further complicate the specification by allowing tunability for what seems already like a complex arrangement of multiple receivers / transmitters … It would seem to me that addition of tunable elements does not improve the yield. I spent two last days reading a bit of latest publications on optical integration and conclusions do go along with what Dekun Liu indicated in the email below ? yield is low and decreasing rapidly as the number of devices in a single package increase. There are some promising integration techniques, but until they become mainstream and we have ability to use silicon-processing-like techniques for optical components, optical integrated components will be more expensive.

 

Perhaps there is something I am missing from your email, but I do not see how tuning helps in this case

 

Marek

 

From: Francois Menard [mailto:fmenard@xxxxxxxxxxx]
Sent: Wednesday, February 17, 2016 8:18 AM
To: STDS-802-3-NGEPON@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_NGEPON] 802.3ca (100G-EPON) bi-weekly consensus - call for agenda items

 

Folks,

 

Please allow me to add some comments to this discussion. I will limit the scope of my apropos to the channel(s) in the direction of the ONU from the OLT.

 

I would not say the situation is as grim as an insertion loss of 4 dB’s for the 4-channel Mux/Demux function requiring AWG, as the insertion loss can be better with more sophisticated integrated optics filters than AWG’s such as contra-directional couplers.  

 

It strikes me that in the 100Gps ONU, what remains a constant source of costs is the requirement for 4 of everything (APD’s, TIA’s, Limiting Amplifiers, Dispersion Compensation) in order to achieve 100Gbps.  It is unclear to me whether APD vendors would be able to cost reduce an array of four APDs, but I suppose anything in volumes of millions is worth its weight melted at the metal yard.

 

My point is that in contrast to NG-PON2,  should the ONU receiver in NG-EPON not be tunable, then NG-EPON ONU’s which will implement fewer than 4 channels, will be hardcoded to certain channels, removing the flexibility of dynamic bandwidth assignment across all NG-EPON channels.  The same of course apply to an NG-EPON ONU cost reduced to a single 25G channel.

 

Another point, is that should the number of channels be greater than four, say 12 or 16 (16x25 = 400 Gbps on the PON), then even if there is 4 of everything in the ONU (APD’s, TIA’s, Limiting Amplifiers, Dispersion Compensation), then ideally, all of the four receivers in the ONU would need to be tunable across the range of 12 or 16 channels.  

 

I would recommend that NG-EPON keeps implementation flexibility to support tunable receivers.

 

-=Francois=-

 

 

 

From: Liudekun <liudekun@xxxxxxxxxx>
Reply-To: Liudekun <liudekun@xxxxxxxxxx>
Date: Wednesday, February 17, 2016 at 05:01
To: "STDS-802-3-NGEPON@xxxxxxxxxxxxxxxxx" <STDS-802-3-NGEPON@xxxxxxxxxxxxxxxxx>
Subject: Re: [802.3_NGEPON] 802.3ca (100G-EPON) bi-weekly consensus - call for agenda items

 

Hi Marek

  I agree with you “the module cost increase is not really linear in the function of integrated channels”

 

Actually , optics integration hasn’t save  cost compared with discrete components by far,  due to the limited yield for optic chips .(This is contrary on electronic chips, VLSI).

A 4 Tx array are more than 4 times expensive than a single channel Tx, due to the yield is the multiplication of single chips’ yield.

For the  package process of optics, especially the coupling, active coupling (need a light source during the coupling) combined with manual adjustment is still the major technology in nowadays ,  so align a 4 channels array to a fiber array are much more complicated  than a single channel alignment .   All of these make the optics integration still very expensive, more expensive than 4 times of single channel. The more channels, the cost becomes even higher.

 

Another aspect we need to consider is that,  if we consider the interface of  a 100G ONU is same with 1G or 10G EPON ONU,  retain 1 fiber , bidirectional,   then we need a 4 wavelength mux and demux inside the module .  the mux and demux will bring extra cost and extra insertion loss.

 

For a 4 channel AWG, the insertion loss is at least 4dB,  if we consider the cascaded  thin film filter,  the IL is still 2~3dB.   If we want to keep a same output power with current 10G EPON ONUs,  then the transmitter will  need to increase the output power by 3dB, the APD receiver will need to improve its’ sensitivity by 3dB,  this will further increase the cost.  (That’s the miracle ,   when the speed comes high ,the sensitivity becomes low,  but we need  the sensitivity becomes even higher )

(The power budget in PON system is always expensive,  if you want to more power budget, then you need to pay more,  because the margin has been consumed by the splitter. So any components with insertion loss means extra cost on the transceiver.)

 

Best regards

Dekun Liu

____________________________________________________

Advanced Access Technologies Dept. 网络研究接入技术部

Huawei Technologies Co., Ltd. 华为技术有限公司 <image001.jpg>
  Phone: +86 027-59267217  Email: liudekun@xxxxxxxxxx

 

 

From: Marek Hajduczenia [mailto:marek.hajduczenia@xxxxxxxxx]
Sent: Wednesday, February 17, 2016 5:42 AM
To: STDS-802-3-NGEPON@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_NGEPON] 802.3ca (100G-EPON) bi-weekly consensus - call for agenda items

 

I would like to add a few more thoughts on individual slides:

 

-          Slide 10: I am not really sure how much power saving there can be in disabling wavelengths versus resulting implementation complexity. Consider that disabling wavelengths in downstream would be rather rare, given that there is always some data to be sent downstream to end customers. Trying to load balance and move all traffic across wavelengths to Lambda0 to be able to disable higher order Lambdas would require much more complexity and control of destination per frame than we have today within an 802.3 PHY.
In the upstream direction, given that ONU powers up lasers only for the duration of the burst, any power savings from powering up one versus 2 or 4 lasers would be also marginal. I think an approach with disabling wavelengths would only work if we had a solid study that demonstrated substantial saving of uW/bit/sec versus implementation complexity.

-          On slide 12, “The cost of 50G and 100G optics will be roughly 2x and 4x of 25G optics” ? I have been looking for actual data to demonstrate the relationship between number of channels and relative device cost and failed to find anything solid so far. I recall discussion in one of multi-lane 100G Task Forces indicating that the module cost increase is not really linear in the function of integrated channels, but I do not see actual materials on this topic. It might be worth to seek contributions from companies with optical integration experience to discuss maturity and also relative cost curves.

-          Slide 4: a lot of good points here.
I am not sure, though, what value would be for “3rd gen ONU can pretend to be either 1st gen or 2nd gen ONU” ? this means that 100G device would operate only as 25G device, which is counterintuitive ? why would someone deploy more expensive device up front and not take advantage of its capabilities? Usually, ONUs are deployed in the function of OLT capabilities, i.e., when the OLT becomes 100G capable, ONUs would follow, not the other way around.
On “3rd gen OLT can serve 1st, 2nd, and 3rd gen ONUs” ? one point to consider is how far this backwards compatibility really needs to go. I am not sure what the right answer is here, because this mostly depends on relative cost of 50G and 100G devices when compared side by side. If the delta is not that big, deploying 100G up front would save on truck rolls and upgrade complexity and perhaps eliminate some extra complexity in the system design ?
“No Rx/Tx tunability” ? we could very much be silent in the spec about tuneability requirements and support is via extensions to MPCP, where the ONU could announce its tuning capability during MPCP discovery and then have channels allocated with REGISTER message. This might open doorway to reuse of some of existing transport grade components, if that makes any economic sense.

 

Marek

 

From: frank effenberger [mailto:frank.effenberger@xxxxxxxxxx]
Sent: Tuesday, February 16, 2016 3:37 PM
To: STDS-802-3-NGEPON@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_NGEPON] 802.3ca (100G-EPON) bi-weekly consensus - call for agenda items

 

Glen and all,

I would caution against thinking that “wavelengths” are like a commodity.  Unlike people, not all wavelengths are created equal, I’m afraid to say. 

Moreover, we don’t allocate individual wavelengths, but actually wavelength bands, with width of which has a huge effect on cost of components.  

This is what the presentations at the last meeting from Ed and I were getting at. 

 

The worst thing is that we are not starting from a clean slate ? there is a lot of legacy there, and also other optics defined in the market that could be reused.

All in all, it is a very complex decision to make, and you can’t simplify it to “Keep the number of wavelengths to a minimum”. 

 

Other than that, your basic ideas of the fully flexible kind of system I generally agree with.  Indeed, my preso in Dallas suggested many of these same features. 

They are a nice ‘wish list’ at this point.  The killer question is: can we achieve all of them?  It’s not so clear to me now.

 

Sincerely,

Frank E.

 

 

From: Glen Kramer [mailto:glen.kramer@xxxxxxxxxxxx]
Sent: Tuesday, February 16, 2016 2:05 PM
To: STDS-802-3-NGEPON@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_NGEPON] 802.3ca (100G-EPON) bi-weekly consensus - call for agenda items

 

Curtis,

 

I'd like to discuss the general architecture approach. We had a number of presentations in Dallas leading towards this approach, but since the Dallas meeting was per-TF, we didn't make any formal decisions. In Atlanta, we started looking into various low-level details, but the big picture is left undefined. My slides are attached. All feedback is welcome.

 

Thanks,

Glen

 

On Mon, Feb 15, 2016 at 10:49 AM, Curtis Knittle <C.Knittle@xxxxxxxxxxxxx> wrote:

 

Dear Colleagues,

 

This coming Thursday, February 18, 11:30-1:00 Mountain Time, we will hold an IEEE 802.3ca 100G-EPON consensus building meeting. Please let me know by Wednesday noon (Mountain Time) whether you have requests for agenda time.

 

If I do not receive agenda requests by noon Wednesday, I will cancel the meeting.

 

Thank you!

 

Curtis

 

 

 

 

Curtis Knittle

VP Wired Technologies ? R&D

CableLabs

desk: +1-303-661-3851

mobile: +1-303-589-6869

c.knittle@xxxxxxxxxxxxx

 

Stay up to date with CableLabs: Read the blog and follow us on Twitter

 

 

--

--------------------------------------

Glen Kramer

Broadband Technology Group

(707)529-0917