Slashdot Mirror
Terabit Ethernet Is Dead, For Now
Nerval's Lobster writes "Sorry, everybody: terabit Ethernet looks like it will have to wait a while longer. The IEEE 802.3 Industry Connections Higher Speed Ethernet Consensus group met this week in Geneva, Switzerland, with attendees concluding—almost to a man—that 400 Gbits/s should be the next step in the evolution of Ethernet. A straw poll at its conclusion found that 61 of the 62 attendees that voted supported 400 Gbits/s as the basis for the near term 'call for interest,' or CFI. The bandwidth call to arms was sounded by a July report by the IEEE, which concluded that, if current trends continue, networks will need to support capacity requirements of 1 terabit per second in 2015 and 10 terabits per second by 2020. In 2015 there will be nearly 15 billion fixed and mobile-networked devices and machine-to-machine connections."
No, unbounded latency. It'll happen, just not yet.
We need terabit Ethernet NOW, not in a decade.
You know my 5 year old nephew keeps confusing need and want too.
How much are you prepared to pay for this desire? If it will cost say 4 times greater per bit to implement Terabit with current technology do you still want it?
"The weirdest thing about a mind, is that every answer that you find, is the basis of a brand new cliche" -
I realised I wasn't being clear about why they can't define the standard now and wait for the technology to catch up.
A standard like this is always a trade off based on the currently available technology, How fast are your analogue transistors, how much processing power do you have to do forward error correction. How fast are your ADCs/DACs to do signal shaping? This determines things like which coding schemes can you use. Also what market needs this and what costs are acceptable, for example DWDM and all the associated costs is perfectly acceptable if fibre is comparatively expensive, however even though in the 90s that would have been the only way to do 10G now we have the capability to do it electrically; designing the spec too soon and guessing is a really bad idea. We don't know how 20nm and lower process nodes are going to behave well enough to predict what their characteristics will be when this technology reaches maturity, to get that wrong is to end up with a standard that either under performs or is over expensive.
Put it another way, the processor architecture you would choose to achieve 80MFLOPS in 1976 is very different from the architecture you would choose in 2006. Telecomms has exactly the same concerns.
"The weirdest thing about a mind, is that every answer that you find, is the basis of a brand new cliche" -
And what exactly is he doing over ethernet that needs that much speed? I'm only just now looking at upgrading our small business network to gigabit. A couple of years ago the cost of a 48 port gigabit switch was pretty high, but now it's very reasonable
which is totally what she said
We need terabit Ethernet NOW, not in a decade.
You know my 5 year old nephew keeps confusing need and want too. How much are you prepared to pay for this desire? If it will cost say 4 times greater per bit to implement Terabit with current technology do you still want it?
I agree with you completely. Learning to seperate our 'wants' from our 'needs' can make all the difference in our 'consumer-driven' lives.
I may 'want' that shiny new car, but I don't 'need' it. If I have a vehicle that meets my needs, I've learned to be grateful for having that. Coveting that 'new shiny' (new car, other person's money/spouse, phone or internet connection speed, whatever it is) can often lead a person down the road to ruin.
In my experience, I know to be happy and grateful for what I have, and don't waste energy on what I don't have (yet). Of about half the people who win a lottery, 5 years later, they end up wishing they'd never heard about the lottery in the first place. Because it irrevocably changed their lives for the worse, and they realize too late that they were happier before they 'won'. Just my two cents.
---------------
I am so smert! I am so smert..., I mean smart! - Homer Simpson
You may discover that you can't have what you want. There are real physical limitations we have to deal with. One issue, with regards to copper Ethernet, that we are having is keeping something that remains compatible with older style wiring. Sticking with 8P8C and UTP is becoming a real issue for higher speeds. At some point we may have to have a break where new standards require a different kind of jack and connector.
Also in terms of "data center only" devices that isn't how things work. You care what data centers use because you connect to them. There can be big advantages in terms of cost, simplicity, and latency, to stick all on one spec. So 40gbps or 400gbps could well be useful. No, maybe you don't see that to your desktop, but that doesn't mean it doesn't get used in the switching infrastructure in your company.
Also each order of magnitude you go up with Ethernet makes the next matter less. It's going to be awhile before there's any real need for 10gbps to the desktop. 1gbps is just plenty fast enough for most things. You can use things over a 1gbps link like they were on your local system and not see much of a performance penalty (latency is a bigger issue than speed in most things at that point). I mean consider that the original SATA spec is only 1.5gbps.
As for 100gbps, it'll take some major increases in what we do before there is a need for that to the desktop, if ever. 10gbps is just an amazing amount of bandwidth to a single computer. It is enough to do multiple uncompressed 1080p60 video streams, almost enough to do a 4k uncompressed video stream.
Big bandwidth is more of a data center/ISP need than a desktop need. 1gbps to the desktop is fine and will continue to be fine for quite some time. However to deliver that, you are going to need more than 1gbps above your connection.
As a parent of a young one I also hear this 500×/day. But what's the cost of "Terabit now and you're safe for a decade" versus "400Gb now, then rewire & replace all your gear in 3-5 years for 750Gb (if there isn't a standards war you have to gamble on), and then do that all over again in another 3-5 years for 1.1Gb"? Because that's the kind of creep we've seen since the early days of token ring and then 10BaseT. Manufacturers certainly want the step-by-step option but the admins and engineers? Not so much so.
Hardly any 10 base T systems bother with the CDMA/CD system that original ethernet had , in fact its more like a serial protocol rather than a broadcast "in the ether" one now. WHy not just give it a new name?
Your nephew is probably about as mature as most geeks.
OTOH. These standards, by sheer fact that they are referencing 1Tbs needs, are most certainly relevant to the backhaul providers and not any normal business outside of that group. Fractional or non-base 10 speeds have been common in those networks since well before the power of 10 thing came about. Once the rest of the technology catches up and makes the power of 10 thing feasible, then the standard "commodity" equipment picks it up (primarily for marketing reasons IMO). Power of 10 is convenient for math reasons, but frequently means absolutely nothing to the backhaul guys (the early adopters).
Those businesses who purchase the regular "commodity" power-of-10 equipment really should be set for a while with the previously commoditised 10Gb links. They are performant, relatively cheap, available, run across the nation, and hard to saturate with the equipment that plugs into either side. I've worked with 8x10Gb multiplexed cross-country low-latency fiber wan links. It is a ludicrous amount of bandwidth unless you are routing other networks like a backhaul provider. I would struggle to name normal businesses which would be unable to use 10Gb links due to a lack of bandwith (for the immediate future). The needs really are different between these markets.
As an aside, fiber may be sold commonly in 100m lengths, but that has nothing to do with the distance the light will work at properly for the speed it is rated. Some fiber / wavelength pairs are only good for a few feet. Others go km, but not with the same NIC, Fiber, Switches, or patch panels. 100m is a really shitty (too short) standard for datacenter use anyways. Frequently, we will get two cages in a datacenter at different times... and they end up farther than 100m apart making copper irrelevant for that use.
Change is incremental like ripples, but big changes come in waves. Back-haul wants the ripples, everyone else wants the wave. I say, let them have their ripples and pay for the development of the waves. It saves both groups of consumers money so long as there aren't TOO many ripples per wave.
- Toast
Yep it's definitely not a technical problem, after all getting serial data to run at 312.5 Gbps over long distances of un-shielded twisted pair copper is simple. The edges of the data are only in the 1.2 THz range after all.
Even on a PCB, 312.5 Gbps gets tricky and expensive, over long distances of fiber or copper it will be very difficult. Dropping to 400 Gbps brings it into the realm of slightly possible but still ridiculously expensive, plus at 400 Gbps you can bond just three links and get 1.2Tbps through, well probably less after overhead.
Damn CS/CE's think they know RF!
Sig is on vacation
"Only high performing computing and Virtualization servers use more than Gigabit links today, but TenGigabit bundles and higher bandwidth links are used on almost every large network on core connections and core to distribution."
I know quite a few non-science professional fields that saturate gigabit to each desktop, and would go for Infiniband, or 10Gig-E if it was viable outside of big corps. Editing/compositing of HD or greater resolution movies shuffle HUUUUGE amounts of data around, and you need a decent turnaround time for the data....