Brad Would you be so kind as to list all the contributions which describe the 20m reach applications, including describing the novel form factors you have alluded to in your earlier email.
Thank you
Chris
From: Brad Booth [mailto:Brad_Booth@xxxxxxxx]
Sent: Tuesday, April 30, 2013 3:09 PM
To: STDS-802-3-100GNGOPTX@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_100GNGOPTX] 802.3_100GNGOPTX : 'Cost power and size differences of proposed MMF PMDs' - seeking supporters/detractors Dan, There were contributions made to the Study Group and Task Force. These resulted in a 20m reach objective. There was a contribution on the issue with module form factors for servers and intra-rack connectivity. You are correct that many of the presentations have been leaning towards optical modules. The concern with Jonathan’s presentation (sorry to pick on you again Jonathan) was that the presentation title was making a comparison of PMDs when in fact it is making a comparison of a specific implementation of an optical module (more explicitly, a CAUI-4 optical module). There is no problem making a comparison a 100m retimed CAUI-4 optical module to a 20m unretimed CAUI-4 optical module, but it is unfair to equate that to being a comparison of a 20m PMD and a 100m PMD. Mike Dudek’s email makes the point: “In my opinion the difference between the optical specs needed for a 20m reach and a 100m reach provides negligible cost, power or size savings in the PMD (changes in laser bias, marginal change in laser driver, or tiny relaxation in receiver sensitivity). I’d estimate the power saving at less than 10mW per port with zero cost or power saving.” That was the PMD specific comparison that the Task Force should consider as it does relate to what P802.3bm will put into the standard. Thanks,
Brad From: Daniel Dove [mailto:ddove@xxxxxxx]
Sent: Tuesday, April 30, 2013 4:17 PM
To: STDS-802-3-100GNGOPTX@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_100GNGOPTX] 802.3_100GNGOPTX : 'Cost power and size differences of proposed MMF PMDs' - seeking supporters/detractors Hi John,
Good observation. We are indeed writing specifications based on the material presented to our Task Force.
As you are aware, we are a contribution driven organization. It should not be assumed by anyone that work will just get done without contributions to drive it. Jonathan's presentation and argument does lean toward optical modules, but then that is where the vast majority of presentation has been given and work has been done.
Dan Dan, Yes I understand your point. However, I will point to the 802.3bm PAR 5.2.b. Scope of the project: This project is to specify additions to and appropriate modifications of IEEE Std 802.3 to add 100 Gb/s Physical Layer (PHY) specifications and management parameters, using a four-lane electrical interface for operation on multimode and single-mode fiber optic cables, and to specify optional Energy Efficient Ethernet (EEE) for 40 Gb/s and 100Gb/s operation over fiber optic cables. In addition, to add 40 Gb/s Physical Layer (PHY) specifications and management parameters for operation on extended reach (> 10 km) single-mode fiber optic cables. Please note no reference to modules – but the reference to Physical Layer (PHY) specifications. So while your point is fair, I will point out, as I always do, we write specifications, not specify implementations, though obviously implementations can influence the specifications. In regards to Jonathan’s email that he just sent – in all fairness this point didn’t really hit me until I saw his presentation, where the term PMD was used to describe the module. Unwrapping the logic here caused me to have this revelation. And that timing caused me to question Jonathan’s statement. The added caveat regarding that the statement is targeting module implementations is correct. Personally, when I supported the 20m objective I envisioned it for high volume applications, where cost for a high volume application needs to be considered, and not just a single module. These are different application spaces with different cost requirements. And Jonathan, I can ensure you that this issue will come up at 802.3, but perhaps not 802.3bm, as I know another study group where I know an author will bring this issue up. J John Hi John,
I always wondered what that indent on your forehead was. I figured it was either an old scar from 802.3z, or perhaps a frying pan? :p
The observation regarding target implementations is valid, however, we should note that since this project began, the focus has been on reducing the power, cost and size of optical modules. See our CFI presentation for more specifics on that.
Its clear that some folks may consider optics directly on board to achieve even higher densities than we are enabling with 4x25G modules, but since that has not been the focus of our work and without additional presentation on that area of concern, we are bound to move forward along the current path.
It would be helpful to receive material on this subject if there is an interest in presenting and we still have a window of opportunity (Friday's the deadline) if someone wishes to submit a request. Key in my mind would be some data on the trajectory for adoption of optics-on-board as opposed to modular optics in the data center environment.
Best Regards,
Dan Dove
Chair, IEEE P802.3bm
On 4/30/13 1:06 PM, John D'Ambrosia wrote: Ok – so thanks for the laugh and pointing out how long I have been at this J The analogy and point, however, is still valid. The impact and assessment of cost is dependent on an individual’s target implementation. This goes straight to the heart of Brad’s point IMHO. I think Mike Dudek’s reply was a very good summary, so rather than ruin his eloquence I’ll refer you to his e-mail - thank you Mike! I should add that the counterboring controversy was well before my time – and I respect you all the more for ‘fessing up to the memory John …. Best wishes J jonathan Jonathan, Can I ask what perspective are you making your statement from regarding sufficient cost, power, or size, to justify another PMD spec. Would you consider it to be the sorts of deployments you see associated with optics today or with high volume manufacturing? I ask this, because I remember back in the early days of Backplane Ethernet when people argued over the cost associated with counterboring. I remember saying many times that a certain company had imprinted the letters HVM into my forehead. John Hi Brad, Thanks for comments ! Ryan and I just compared the proposals for MMF PMDs that have been contributed to 802.3bm, with (for myself at least) a very module-centric viewpoint, to see how our project goals of significant reduction of cost, power, or size, would be addressed by the un-retimed PMD proposal. Slide 2 was intended to provide a short summary of the context of the comparison, it refers to the 100 m reach MMF PMD baseline proposal, which makes it clear that CAUI-4 is an example (not a required) 4 lane interface. I’m happy to add an explicit reference to that baseline if that helps ? Whether the implementer uses CAUI-4 or not, for the PMD specs to be viable, modeling shows that there are very tight jitter requirements at the PMA/PMD interface; putting retiming inside the module is one way of making that so (for the 100m reach baseline). I agree there may be many proprietary methods for meeting the optical specs for the 802.3bm 100m MMF PMD – but Ethernet doesn’t disallow any of those by not standardizing an un-retimed interface - it is a given that proprietary solutions meeting the PMD specs are allowed and expected. But if even they could allow use of a module without retiming, it’s not going to save sufficient cost, power, or size, to justify another PMD spec. Best wishes jonathan Jonathan, I want to touch on your statement that the “PMA is expected to be inside the module.” That is my primary concern. IEEE P802.3bm does not specify modules, nor does it specify implementations. CAUI-4 is an OPTIONAL instantiation of the PMA service interface. The use of CAUI-4 as the interface to an optical module or for any other means is an implementation decision. While the presentation is correct relative to a very specific implementation using CAUI-4 optical modules, there is a potential market that is being ignored by assuming they will use CAUI-4 and/or an optical module. Would you (and your supporters) be willing to add clarification to the presentation that the material assumes the use of a CAUI-4 optical module and other implementations could result in different cost, power and size estimates? Thanks,
Brad Hi Brad, Thank for your comments. You are right, the clock and data recovery are part of the PMA; for the 100G 100m MMF baseline that part of the PMA is expected to be inside the module. You raise an interesting point about having different power settings controllable by management software – my initial thoughts are that the difference between 20m and 100m reach link budgets is about 1.4 dB, it’s probably not very significant in terms VCSEL bias current savings for example. I don’t think CDRs could be turned off without tighter jitter specs being set at the electrical interface - as some of John Petrilla’s work has shown in the MMF ad hoc meetings, the chip to module CAUI-4 specs don’t support an un-retimed optical link. John’s work showed that 20 m reach PMD optical parameters would need to be as tight as the 100m baseline in order to make the jitter specs achievable. Of course, I’m happy to hear other opinions Best wishes jonathan Jonathan, Thanks for forwarding the presentation. There is one aspect that does concern me as I reviewed the material and that is it is only making a comparison of unretimed vs. retimed implementations. As far as my understanding of a PMD and its PMD service interface, there are no clock and data recovery circuits (see Annex 86A). Therefore, I think the presentation has not really focus on the differences that could be specified relative to the PMDs. I’m not an optics expert, so please feel to correct my following interpretation/thought-line. It would seem to me that the Task Force could specify different optical power requirements (one for 20m and another for 100m) that could be controlled via the management interface. The 100m power level would be the default, and the 20m power level could be selected by either management software or the end user. Would this scenario offer any potential power savings? Thanks,
Brad From: Jonathan King [mailto:jonathan.king@xxxxxxxxxxx]
Sent: Friday, April 26, 2013 6:32 PM
To: STDS-802-3-100GNGOPTX@xxxxxxxxxxxxxxxxx; a_flatman@xxxxxxxxxxxxx; Abbott, John S Dr; Amezcua, A. (Adrian); Andy Moorwood; Anslow, Peter; Anthony Torza; Bernstein, Gary; Booth, Brad; Daniel Dove; Ephrem Wu; Gary Nicholl (gnicholl); Harry Fu; Jack Jewell; Jeffery Maki; John Petrilla; Keith Nellis; Kolesar, Paul; Lian Zhao; Martin Gilpatric; Phil.McClay@xxxxxx; Mike Peng Li; mike.dudek@xxxxxxxxxx; mnowell@xxxxxxxxx; Oren Sela; Piers Dawe; Rabinovich, Rick (Rick); Rick.Pimpinella@xxxxxxxxxxx; Robert Coenen; ryan.latchman@xxxxxxxxxxxxx; Scott Kipp; Shmuel Levy; Swanson, Steven E; Tracy, Nathan L; Vipul Bhatt
Subject: 802.3_100GNGOPTX : 'Cost power and size differences of proposed MMF PMDs' - seeking supporters/detractors Dear all, Following a couple of comments, I’ve revised the slides slightly to be more clear that those listed as supporters (or detractors) are supporting the conclusions, rather than necessarily 100% agreeing with the specific details in the presentation . Obviously power estimates vary somewhat between companies. I’ve moved the supporters and detractors lists to the end of the presentation and labeled them as ‘supporting the conclusions’ and ‘not supporting the conclusions’ . Best wishes jonathan From: Jonathan King <jonathan.king@xxxxxxxxxxx>
Date: Thursday, April 25, 2013 1:44 PM
To: <STDS-802-3-100GNGOPTX@xxxxxxxxxxxxxxxxx>, Jonathan King <jonathan.king@xxxxxxxxxxx>, <a_flatman@xxxxxxxxxxxxx>, "Abbott, John S Dr" <AbbottJS@xxxxxxxxxxx>, "Amezcua, A. (Adrian)" <adrian.amezcua@xxxxxxxxx>, Andy Moorwood <amoorwood@xxxxxxxxxxx>, "Anslow, Peter" <panslow@xxxxxxxxx>, Anthony Torza <anthony.torza@xxxxxxxxxx>, "Bernstein, Gary" <GBernstein@xxxxxxxxxxx>, Brad Booth <Brad_Booth@xxxxxxxx>, Daniel Dove <ddove@xxxxxxx>, Ephrem Wu <ephremw@xxxxxxxxxx>, "Gary Nicholl (gnicholl)" <gnicholl@xxxxxxxxx>, Harry Fu <harryf@xxxxxxxxxx>, Jack Jewell <jack@xxxxxxxxxxxxxx>, Jeffery Maki <jmaki@xxxxxxxxxxx>, John Petrilla <john.petrilla@xxxxxxxxxxxxx>, Keith Nellis <knellis@xxxxxxxxx>, "Kolesar, Paul" <PKOLESAR@xxxxxxxxxxxxx>, Lian Zhao <lzhao@xxxxxxxxx>, Martin Gilpatric <marting@xxxxxxxxxx>, <Phil.McClay@xxxxxx>, Mike Peng Li <mpli@xxxxxxxxxx>, <mike.dudek@xxxxxxxxxx>, <mnowell@xxxxxxxxx>, Oren Sela <orens@xxxxxxxxxxxx>, Piers Dawe <piers.dawe@xxxxxxxxxxxxx>, "Rabinovich, Rick (Rick)" <rick.rabinovich@xxxxxxxxxxxxxxxxxx>, <Rick.Pimpinella@xxxxxxxxxxx>, Robert Coenen <rcoenen@xxxxxxxxxxxxxxxxxxx>, <ryan.latchman@xxxxxxxxxxxxx>, Scott Kipp <skipp@xxxxxxxxxxx>, Shmuel Levy <shmuel@xxxxxxxxxxxx>, "Swanson, Steven E" <SwansonSE@xxxxxxxxxxx>, "Tracy, Nathan L" <ntracy@xxxxxx>, Vipul Bhatt <Vipul_Bhatt@xxxxxxxx>
Subject: 802.3_100GNGOPTX : 'Cost power and size differences of proposed MMF PMDs' - seeking supporters/detractors Dear all , Ryan and I are seeking supporters and detractors for the attached joint presentation: “Cost, power, size differences of proposed MMF PMDs: 20 m reach un-retimed vs 100 m reach retimed baseline”, which will be submitted for to the May meeting of 802.3bm in Victoria, BC. The slides conclude that “• By about the same time 802.3bm is technically stable (H2 2014) there will be no significant power, cost, or size, advantage to be gained from an un-retimed short reach PMD. – A 20 m reach un-retimed PMD would not meet the criteria for distinct identity. • The 20 m reach objective is met by the 100 m reach PMD, a separate 20 m PMD is not required” If you agree (or disagree) with the presentation and wish to be listed as a supporter (or detractor) of the presentation, please let me know by e-mailing me. Thanks ! Best wishes jonathan jonathan.king@xxxxxxxxxxx
|