For more than a decade, Ethernet and optical transport has converged around a common set of rates (10G, 40G, 100G) which have facilitated interconnection of Ethernet over transport networks and advantages of common components for various applications. Evolving beyond these rates, modern optical networking gear based on complex, polarization multiplexed modulation formats with intradyne, coherent receivers tend to be DSP-based implementations which can often be programmed to support a variety of different modulation formats where the complexity of the constellation (number of bits per symbol) and the symbol rate can be selected based on the reach required for a particular link. There has been considerable discussion about how to provide efficient Ethernet-based router to transport interconnection in this more flexible optical networking environment.This panel brings together a group of distinguished experts representing cloud-scale data center operators, router manufacturers, optical networking equipment manufacturers, and component vendors. The panel will explore the merits of using flexible rates of Ethernet for router to transport interconnection, including configurations, use cases, and the implications for routers and transport gear.
All of the abstracts are up on line at the Ethernet Alliance website for your review.
This is a very nice effort of aggregation on technical front, but is it not possible to have 10Mbps to 1000Gpbs common interface and based on the initial negotiation of both the parties with link speed will be determined using Flexi MAC Sublayer? This will be applicable for all the new Ethernet adaptors.
And, as I suggested quite recently, the same debate/question will no doubt occur with respect to 400G Ethernet, where 500G would be more in line with Ethernet traditional bit rates (in that it aggregates to a powers of 10 bit rate, which has always been Ethernet tradition), and performance-wise, well, what difference could it make?
The only valid excuse for the 4x oddities in the Ethernet lineup, only two of them in fact (40G and 400G), was simpler carriage over non-Ethernet physical media. If that non-Ethernet physical medium is either not used, or never existed at all (in case of 400G), why would these two rates survive? Once you have the 10/25/50 options available, one would expect the oddballs to be dropped along the way.
I don't think any of this is terribly earth shaking, mind you. Ultimately, users couldn't care much less, as long as the equipment is available at a decent price. This is mostly about tradition, IMO.
@Bert: Good points. I sense there is already concern about having fractured the ecosystem.
Companies need to focus their investments. They once thought 40G made sense but now some are wondering if they can afford to make chips, cards, sw for 40G AND 50G.
There is some clear data center support from Google and Msoft as end users and their vendors for 25/50 now for a handful of reasons. As you point out that might push for a 500 not 400G standard, but the 400G effort is already embarked. Hmmmmmm
The title of your post is dead-on - after our last exchange it is totally what i expect. While you are focusing on individual rates, your comment regarding x4 oddities is so not in sync with today's Ethernet environment. 40G is not going to disappear simply because we now have 25 or in the future 50G. Why? Because 4x10 GbE is enabling hhigher density solutions of 10G that before.
Also, if you go and look at the IEEE 802.3 Ethernet specifcation - you will mind multiiple instances of x4 configuration within the specificaiton - XAUI, Backplane, CX4, BASE-T, new 4Pair PoE, 100G, and now 400G.
I am guessing from your prior comments that you have more of a carrier emphasis. The Ethernet Alliance TEF is a great event to come and hear from different segments and learn what is really happening in each of the different segments. I would recommend you consider attending. SOme of the things you have said indiciate that this event would help provide you with some additional insight.
I would agree that an existing standard, such as 40G, will not disappear. But a new standard, such as 400G, may never come to fruition. As you said before, there was a lot of initial opposition to the 40G standard, so now that we're debating 25 and 50 bit rates instead, my bet is that the emphasis on anything based on 40 will diminish.
The 4-lane Ethernets of today, that you mentioned in your post, aggregate to a powers of 10 bit rates. That was my point. That's why these 4-lane solutions use 2.5 or 25 or 250, not 4 or 40. For a 4-lane Ethernet to aggregate to anything else, like 400G, would be the novelty. I'm not sure why 802.3 membership would be pushing hard for such a choice. My bet is, they won't.
I'm not sure about the density argument. If 4*10G is high density, then wouldn't 5*10G be equally high density, and 2*25G be even higher density? We have to see what the vendors come up with, but my suspicion is that either 2*25G, or 1*50G, will beat out 4*10G, in terms of density, for anything in the 40G-50G speed range. And then you can aggregate these easily into 100G or eventually 1T bit rates.
As to physical layer signaling, I'm not sure why that would be dramatically different, between, say, the various PMD/PHY used for 40G, and new ones for either 25G or 50G.
The carrier arguments, i.e. carriage over SONET/SDH, were only to show that *a* motivation for 40G was that. A pre-existing SONET standard, available since 2001, immediately available for metro area Ethernets. But that rationale does not hold for 400G. So if the membership of IEEE 802.3 checks the potential SONET/SDH rates of the future, if that even matters anymore, they will notice that SONET/SDH will jump from 39.813 Gb/s to 159.252 Gb/s to 637.009 Gb/s, if we continue to increase the rate using the same sequence used to date. There is no 400G step in that sequence. Which means, you won't be using the entire concatenated SONET/SDH frame at 400G or 500G rates anyway. You'd have to use virtual tributaries to achieve such intermediate rates over SONET/SDH, and either 400G or 500G will just as easily pack into such VTs.
So bottom line, I'm not saying that SONET/SDH rates matter greatly anymore, nor am I a big proponent of such schemes. I am saying that SONET/SDH can't be used as *a* rationale for 400G, and do so in opposition to 500G, as it was at 40G. It's just as easy to easy to pick 500G as it is to pick 400G, if you insist on carrying these Ethernets over SONET/SDH.
As we discussed before - Sonet was not a justification for 40GbE, you just refuse to acknowledge what i have told you.
Not all of the 4 lane rates i mentioned do not aggregate to powers of 10 - 40GBASE-nR4. and your point about 5 lanes of 10G is very easy to dismiss - no one would do it. We talked about this in relation to 100G and doing 5 lanes of 20G - why didnt we do it- because we like to work in powers of 2. It is a physical thing, not an aggregate rate in this instance.
You are proving my point about the TEF and this is all part of the rate debate. Oh so 50G is next lane rate - so we should do 4 lanes of 50 or 200G which is then an upgrade from the same form factor for 40G and 100G.
Now in regards of 4 lanes of 10G getting beat out by 2 lanes of 25G or 1 lane of 50G - i wish it were that simple. The reality is that the solution for the application will be dictated by the economics of the application. FOr example, while cloud scale data centers are very interested in 25G, the bulk of servers being shipped today is 1G, and not for cloud scale. You have to take the timing of the market you are targeting into consideration, and guess what - differnet markets in a very real way.
I wasted a lot of time trying to post replies with links in them, forgetting that EE Times is filtering those posts out. Oh well.
As we discussed before - Sonet was not a justification for 40GbE, you just refuse to acknowledge what i have told you.
Yes, because I have found any number of online links, even from within IEEE 802.3, even from EE Times itself, dating back as far as 2002, which say otherwise. The Metro Ethernet Forum *is* interested in OTN carriage of Ethernet, and it's been one aspect that's been with Ethernet ever since the 10G days. Too bad links don't work on these posts. Although as I said, SONET/SDH per se are no longer being considered, past 40G rates. Now we're talking about Optical Transport Units (OTU), as opposed to STS or STM of SONET/SDH. Similar concept, but more flexible. (I'm not a telco type, btw.)
We talked about this in relation to 100G and doing 5 lanes of 20G - why didnt we do it- because we like to work in powers of 2. It is a physical thing, not an aggregate rate in this instance.
I can't parse this. Just as 4 lanes of 250M work for 1G, four lanes of 25G can work just fine for 100G, as could two lanes of 50G. I don't get this powers of two comment at all. Ethernet went from 10 to 100 to 1000 to 10G, both as single lane and as 4 or more lanes, and I see no rationale for saying this can't go on. I'll bet that in these new discussions wrt 25G and 50G, you'll see vendors stating this verbatim.
Oh so 50G is next lane rate - so we should do 4 lanes of 50 or 200G which is then an upgrade from the same form factor for 40G and 100G.
I can't parse this either. I could equally say, oh, so now we should be considering 160G and 320G Ethernets, based on multi-lane 40G? There are any number of multi-lane examples that aren't used. Do you know of any 750 Mb/s Ethernets, for instance? Or 1.25 Gb/s variants?
Even though you failed to appreciate my previous mention of this, use of many parallel lanes becomes a nuisance. For timing sync, for connectors, and for cabling. So if you have a practical, cost-effective way of achieving 100G with one or two lanes, it is preferable to using four or 10 lanes. Ask your vendor members if they don't agree. And by the way, this is a general fact, applicable to any interconnect, not just Ethernet. It's ALWAYS a back and forth see-saw, between serial and parallel.
The market will develop whatever there is demand for. There's nothing sacrosanct about 40G or 400G, that vendors can't redesign into 25G, 250G, 50G or 500G, if there's demand for these speeds.
Aw i am feeling the channeling of Bob Metcalfe and feeling snarky!
Yes, please go back and look for these URL's - chances are you will find if you back to 2006 or later - they occurred in efforts that i helped chair or probably even wrote or co-authored. And going back to 2002 from an industry standardization effort is meaningless. I have seen rooms change on the time overnight, so 3 or 4 years are not really of significance to me- a bump in the road if you will.
Now it seems throught your arguing with me - that perhaps you are starting to grasp the debates and arguments that are actually happening in the industry TODAY. This is why the Ethernet Alliance is hosting this forum. Now your question would we consider doing 160G if we have a 40G lane rate - the answer is absolutely yes. But the real issue is that now that 25GbE standardization is underway - is 40GbE or 50GbE the next logical step? People are pointing to 50GbE because it would be a 2x jump over 25GbE, but some are also pointing out that those parts designed to support 40GbE could benefit from going to a serial optical (or cu for that matter) solution, which may or may not work back into the ASIC.
And that is a great thing- perhaps todays serdes that run from 10Gb/s to 25Gb/s could be expanded to 40 or 50Gb/s. Perhaps with MLG we could use the higher rate serdes with glue logic to support higher port counts of lower speeds?
Are you able to parse this?
And i am fully aware of the impact of wide interfaces. THis was an argument we made from a system level perspective by calculating the number of pins to support capacities based on different I/O speeds. THe impact is frigthening. Matter of fact if you look at Ethernet x16 is the widest we ever went. the CAUI-10 based on 10x10 was short-lived, but served its purpose as the world is now progressing to a 4x25G model.
And if you are paying attention to the new group i am chairing - the IEEE 802.3bs 400GbE Task Force - you will see there is a huge debate underway whether the interfaces should be 4x100 or 8x50 optically, as the group also looks at developing 16x25G and 8x50G electrical itnerfaces?
And finally- yes you are right - people can do whatever they want. But that is a debate about the role that standards place in today's world. And that is probably longer than i feel like responding to right now. I will leave you with this old favorite though - DON'T BET AGAINST ETHERNET!
Thanks for your question. This needs a bit of explanation - it was realized with 25GbE that the best cost optimized solution occurs when the Ethernet Rate matches the basic capability of a single I/O into a chip. It reduces CAPEX and OPEX. With today's 40GbE solutions out int he market - you can achieve it in a standard mode by 4x10 or a proprietary mode with 2x20. It is not cost optimal. SO the question is being asked what will the next I/O rate be developed - serial 40G or serial 50G?
Further people are asking if they jump to 25G is 40G which is less than a 2x improvement really the right next step?
Ultimately i think the market will see both solution sets, and it will dependent on invididual customer upgrade paths and timings.
thank you John...exciting battlefield for IC Com vendors!...would you have interests in giving a talk on this topic at emerging technologies symposium in Vancouver in 2015? program at www.cmosetr.com, Kris
Also suggest you consider coming to the Ethernet Alliance Technology Exploration Forum on Oct 16 @ Santa Clara Convention Center. These topics will be discussed in greater detail there by meeting participants.
thank you John...but I no longer work in networking, I ended up my Com IC carrier desiging 10 Gb/s transceivers 10 years ago...still follow up the topic though, this 50 Gb/s vs 40 Gb/s debate is really interesting, I completely didn't see this coming 10 years ago (when 40 Gb/s standard was already in place)...Kris
What are the engineering and design challenges in creating successful IoT devices? These devices are usually small, resource-constrained electronics designed to sense, collect, send, and/or interpret data. Some of the devices need to be smart enough to act upon data in real time, 24/7. Specifically the guests will discuss sensors, security, and lessons from IoT deployments.