back to article Network boffins say Terabit Ethernet is TOO FAST

The IEEE consensus group in charge of developing future networking standards has some words for anyone who was hoping to see Terabit Ethernet in the next few years: not so fast. In July, the IEEE Ethernet Bandwidth Assessment ad hoc group issued a report stating that global demand for network bandwidth is growing at such an …

COMMENTS

This topic is closed for new posts.
  1. Anonymous Coward
    Anonymous Coward

    Program more efficiently.

    Get rid of the crap / spam / garbage we are subjected to daily.

    Then a DSL will be fast enough for everyone.

    1. Greg J Preece

      See that, flying way over your head? That was a clue.

      Improve coding, reduce network usage by 20%.

      Number of users doubles.

      Average media filesize doubles.

      Whoops! There goes your utopia.

      1. Anonymous Coward
        Anonymous Coward

        Not to mention...

        640 K ought to be enough for anybody.

        1. ThomH

          Re: Not to mention...

          Given that there's a world market for about five computers, networking hardly feels essential at all.

    2. Anonymous Coward
      Anonymous Coward

      Have you ever thought that this sort of thing may be for huge corporate networks or WAN links carrying the traffic of thousands of users?

    3. CheesyTheClown

      Wow! Genius!

      As a network hardware designer as well as a network engineer, I am so happy you solved all my problems!

      Now, if I can suggest to my customers that they stop collecting detailed scientific data, stop trying to map genes, stop trying to broadcast video from UN council meetings, stop monitoring seismic activity, we'll be better off!

      Next time, think first.

  2. Zebulebu
    Headmaster

    What, pray tell, is an RJ-45 cable?

    http://www.theregister.co.uk/Design/graphics/icons/comment/headmaster_32.png

    Do you use it with a CAT5 connector?

    1. Anonymous Coward
      Anonymous Coward

      Re: What, pray tell, is an RJ-45 cable?

      Probably meaning any sort of RJ-45 terminated cable, whether it be CAT5, CAT5e, CAT6...etc.

      1. johnnytruant
        Coat

        Re: What, pray tell, is an RJ-45 cable?

        RJ-45 is a cable specification, which defines both connector geometry and the connections thereto. The plug on the end of your network cable is designated 8P8C. When that connector is wired up with CATx grade wire using Ethernet pinouts, the whole assembly becomes an RJ-45 cable.

        Mine's the one with the crimping tool in the pocket.

        1. Alan Thompson
          Coat

          Re: What, pray tell, is an RJ-45 cable?

          Actually, RJ-45 is a Jack connector specification and 8P8C is the equivalent Plug specification while CAT5, CAT5e, CAT6 and CAT6a are all cable specifications that require the use of their own respective versions of the RJ-45 and 8P8C connector - including shielded and unshielded variations.

          Mines the one with the clue in the pocket - you can borrow it.

    2. HMB

      Re: What, pray tell, is an RJ-45 cable?

      @Zebulebu

      I'm afraid I do imagine you as the comic book store owner from The Simplsons.

  3. Christian Berger

    Well, the point is...

    Since you can do bonding, will faster Ethernet be cheaper than just bonding together multiple slow links. Additionally you can then use wavelength multiplex to get up to about 100 wavelength onto a single fibre. If the equipment to run 10 100 GBit connections over a fibre is cheaper than a Terabit line, people will bond the 100 GBit lines.

    1. Anonymous Coward
      Anonymous Coward

      Re: Well, the point is...

      Except that there are limitations into bonding.

      1) Most equipment either supports a maximum of 8 or 16 in a bonded connection.

      2) Most equipment has a limit on the number of bonded links as well.

      3) You need a better ether-channel hash as well. Bonding does not mean full utilization like a single physical link of the same speed.

      4) If you need to use 20 strands of fiber rather than just two, that is a huge expense as is ten times the ports and ten times the optics.

      1. Christian Berger

        Re: Well, the point is...

        1) and 2) That's just a limitation of the current chipsets, maybe even just a software limitation. There are many ways around that.

        3) That depends on the way you bond. You can simply use round robin.

        4) Well that's the point, is the cost of fiber and optics lower than the one of exotic electronics, and believe me, electronics that can handle Terabit streams is still fairly exotic. You can even reduce the amount of fiber by using WDM systems. Besides Terabit Ethernet would most likely use some form of WDM anyhow, so you probably don't save a lot on optics there.

        At the very least, we can take a break and look at the problems we face with 100 GBit Ethernet. Then we can see what problems 1 TBit Ethernet needs to solve.

  4. Mage Silver badge

    Bonding

    That's why they are not doing 1Tbps. It has to be cheaper than bonding (i.e. cost per bit must be lower) since it's fibre only and many fibres in one cable is no big deal.

    So they must think 400Gbps optical can be be cheaper than 2 x 100Gbps. In a reasonable time frame. Since the 1 out of 62 was only voting to decide later, the 400Gbps is pretty unanimous.

    "It's Ethernet, but not as we know it, Jim."

    400GHz happens to be around the upper limit of currently available Microwave ICs. The folks using these buy bare chips rather than fully packaged parts. Perhaps there will be Soc pairing with a laser diode /photo diode interface?

    1. Blitterbug
      Devil

      Re: Fibre ftw...

      ...its spelling, that is. Thanks, Mage, for getting it right. Shame the article itself, on a UK site, couldn't even manage that.

  5. Oscar

    I rememebr the day (back in 2002) when my bosses set up a 1Gbps backbone on their network and wired me and my console dev kits directly on to that network ... I was such a happy bunny. Give me an excuse (and the bandwith) and I'm pretty sure I can properly whore 1Tbps too :D

  6. another_vulture

    trivial

    Today, in the data center we have 10Gbps over copper to each server. An individual server doing a non-trivial job might support a 10GB load. By Moore's law, an individual server may need to support a 100Gbps load in 2020. But we can do 100Gbps today, using WDM, on a single fiber, to each server. So today, CPU power is the constraint, not NIC bandwidth.

    By 2020, a WM NIC should be able to handle 10x110 0Gbe or better, cheaply, on a single fiber. NIC BW will not be the conatraint. LAN BW will follow suit. WAN BW will (again) be the bottleneck.

  7. Anonymous Coward
    Anonymous Coward

    The hardware manufacturers won't allow it...

    ,,, since they're still tryimg to screw as much money out of their customers with 1Gb/s even before they move to 10Gb/s..

  8. Mikel

    Retail specs and standards

    It's important to note that these standards are for retail products that must be universally compatible. If you own both ends of the fiber and can engineer and manufacture the switching technology like Google can do, then terabit IP networking is available now. It's not standard and it's technically not "Ethernet" but it puts the bits down the pipe.

    1. CheesyTheClown

      Re: Retail specs and standards

      Sorry I thumbs downed it, unintentional. My thumb was to big or the little button. I meant to press reply.

      I know Google has terabit links, but not likely over a single fiber. If you consider the requirements, current ASIC technology can't support the speeds being suggested. Consider that a terabit would require possibly a bus that is 1000 bits wide at 1Ghz clocks. With modern chip packaging, that would require a chip with nearly 10,000 pins on a circuit board of more than 50 layers to fan out and shield as a parallel bus. The alternative would be a 100 lane PCIe bus, that would require a chip with 4000 pins on a 40 layer board to accomplish. Mind you, these are rough estimates. PCIe is point to point, not a bus. So you'd need 100 PCIe 3.0 switches to distribute the 100 gigabytes of data across multiple ASICs for processing.

      The alternative would be a single chips with multiple terabit transceivers to handle MPLS switching (as routing at those speeds would be beyond feasible) even a hand laid out internal multiplexer would consume so much chip space that 22nm tech might not be able to handle it. In addition, you'd need enough internal buffer to handle it since external RAM would be out of the question. Even if the chip was designed entirely by Intel, IBM or TSMC, it is highly unlikely that terabit would be achievable on even next gen chip tech.

      So, Google is almost certainly doing something different than what you are thinking.

      1. Steven Roper
        Thumb Up

        @CheesyTheClown

        You know you can change your downvote to an upvote simply by hitting the thumbs-up, right?

        1. Anonymous Coward
          Anonymous Coward

          Re: @CheesyTheClown

          Really?

        2. Anonymous Coward
          Anonymous Coward

          Re: @CheesyTheClown

          I just tried it on your post and correct you are. Except you can't change a thumbs up/down to no thumb at all. So Maybe Cheesy felt that a thumb-down was preferable to a thumb-up given a no-thumb was no longer an option.

          1. CheesyTheClown

            Re: @CheesyTheClown

            Nah.. For Karma's sake and that no one should suffer due to my fat thumbs, I was happy to change it... Just really forgot to try after rambling for so long haha

  9. Semaj
    Unhappy

    I'd just be happy if I could download a file and watch 1 Youtube video at once...

  10. Nigel 11
    Boffin

    Physics LImits

    They're getting quite close to the physics limits. The frequency of light (visible light) is in the range 400 to 800 Terahertz. So 1THz modulation is a significant fraction of the carrier frequency, and dispersion effects will start to be very significant.

    It was solved with radio waves a long time ago, but we don't yet have anything like the control over individual cycles of optical light that we do over individual cycles of RF and Microwave radiation. As far as light is concerned, we're still doing the equivalent of AM radio.

This topic is closed for new posts.

Other stories you might like