| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
Brad Very nicely perspective! The problem with the greater than >10m MMF objective is that we do not have, and will not have in the time frame of the TF, solid data to make the quantitative trade-offs that you outline. The 50m objective has no particular cost basis other than it’s longer than majority of copper links and most HPC links. If that was really the application space that we wanted to support with a second PMD at the lowest possible cost, then 50m is too long; 30m reach is better from a purely application space optimization stand point. Chris From: Brad Booth [mailto:Brad_Booth@xxxxxxxx] I worry about the idea of optimizing for length. In hindsight, that’s what the 300m OM3 solution was when we did 10GBASE-SR. That also seemed to be the underlying theme in the 10GBASE-LRM effort. Both those were in some way burdened by their reach. It seems wiser to remember that we are trying to work with Ethernet, and in the Ethernet market – the volume Ethernet market – cost is king. In the end, does the customer really care about whether it is MMF or SMF used in their data center? Or do they just care about the TCO/ROI equation? I would postulate that it is the latter. So, are we picking 150m OM4 because that’s the lowest cost implementation, or are we picking it because that’s what we were able to do with 10G VCSELs; therefore, we need to support with 25G lasers? Why are we picking 50m OM3? Is there a significant cost difference (3-4x) from the PMD that does Xm on OM3 using the 150m OM4 lasers? What would be the cost difference for a 150m OM4 solution vs. a 150m SMF solution? How about the following: · MMF that does at least 10m. · MMF that does at least 100m. · SMF that does at least 500m. Cheers, From: Piers Dawe [mailto:piers.dawe@xxxxxxxxxxxxx] I also see it as inevitable that there will be two variants: "The first PMD would be optimized for lowest cost at maximum length; the second PMD would be optimized for maximum length at lowest cost." Although 802.3 could ignore one, I don't think the market will. It's worth discussing whether we expect these two things to be interoperable. They might not be PMDs, they could be the same PMD with different PMAs (retimed or not) or FEC. However, we know we don't know accurately enough how a staircase of increasing features/cost/power in the PHY relates to length. Any objective for a particular length now is making a resolution that some particular length is more important than cost, power, size, or anything that is not an objective. It is so important that it must be the answer, whatever the technology can deliver. It is making a technical decision in the Study Group that the Task Force should make. Much worse, it is making that decision when we know we don't know what it means (e.g. will it require faster lasers? FEC? DFE in the module?). The length has to be a consequence of finding which step in the staircase of more complicated and expensive and power-hungry PHYs is too much, and then what reach can be delivered in the future with acceptable yield, before that step. Steve's words point out what a more sensible pair of objectives would include: one optimized for length; one optimized for cost and power. I have to take issue with Jack's "trend". As I have said before, lasers don't follow Moore's Law; we are discussing different PHY architectures not a constant architecture but with faster silicon. So we can't set objectives recklessly and rely on time to heal the wounds. Piers From: Jack Jewell [mailto:jack@xxxxxxxxxxxxxx] I support standardizing these 2 Objectives. The first is a slam-dunk lowest-cost solution (no electronic "frills") for growing-and-important HPC and copper-replacement applications. The second takes into account the general trend that well-targeted objectives-set-now may appear hard now, but become practical by the time volume markets mature. Cheers, Jack From: "Swanson, Steven E" <SwansonSE@xxxxxxxxxxx> Dan etal, I also support Jonathan’s effort to conduct the a straw poll on the MMF objectives and have responded to Jonathan with the same proposal that I sent to the broader reflector last week, namely: Define a 4-lane 100 Gb/s PHY for operation over OM3 MMF with lengths up to at least 50 m Define a 4-lane 100 Gb/s PHY for operation over OM4 MMF with lengths up to at least 150 m I wanted to provide some more rationale for why I chose these two objectives as opposed to the 100m objective on MMF that so many seem to favor. I could not support a 100m objective over MMF because I don’t believe that it optimally addresses either of the two customer requests that I have heard since the formation of the SG. We have to remember that the goal of this SG is to establish objectives leading to cost effective 100G solutions. I believe the current solution, 100GBASE-SR10 (supporting 100m on OM3 and 150m on OM4) is cost effective and that is supported in that we are using it as the baseline for comparison of all other PMDs. The current estimates for 100GBASE-SR4 have typically come in at 1.2x 100GBASE-SR10 providing further support for its cost effectiveness. We believe, however, that long term that 100GBASE-SR4 will be more cost effective than 100GBASE-SR10. So the real question is what link length objective(s) should we set for 100GBASE-SR4? I would argue that it is not 100m. I don’t believe that we can get to 100m with the lowest cost transceiver, thereby requiring those customers who do not need 100m to pay for something they don’t need. Our modeling suggests that we can’t get to 100m on OM3 or OM4 fiber with the current transceiver design without adding cost. I think we can support 50m without adding cost. Second, 100m does not meet the minimum link length requirement for a significant percentage of data center customers, thereby denying them a standard solution even if that solution is much more cost effective than a SMF solution that will support the same length. We get calls on a regular basis to support link lengths longer than what is specified in IEEE; we recently revised to 10GBASE-SR spec to include OM4, achieving longer link lengths. Finally, we need a solution that at least supports what we just specified in 100GBASE-SR10 that supports 100m on OM3 and 150m on OM4. I believe an objective that supports 150m on OM4 will support 100m on OM3. We can continue to analyze models and gaze into crystal balls, but I don’t think it will change the answer much. We should set the objectives to address our customer needs and then determine how to meet those objectives in the Task Force. So I agree with Petar Pepuljugoski here - the users should speak up on this. I believe that there is merit for two PMDs for MMF. The first PMD would be optimized for lowest cost at maximum length; the second PMD would be optimized for maximum length at lowest cost. Best regards, Steve Steven E. Swanson t 828-901-5328 From: Daniel Dove [mailto:ddove@xxxxxxx] Jonathan, I support Pete and am also in favour of setting the objective to be 100m over MMF and leave the technical choices as to whether this means over OM3 or OM4 etc. to the TF. Regards, Alan
|