IEEE taking input on next-gen Ethernet standard

The Ethernet standard currently tops out at a staggering 100 Gbps, which requires four 25 Gbps lanes. That’s apparently not fast enough for the folks who build servers, data centers, and high-performance computing solutions, because the IEEE is inviting the industry to weigh in on the next version of the networking spec. The standards body is forecasting bandwidth demands in 2015 to be 10 times higher than what they were in 2010 and another order of magnitude higher in 2020. No wonder a faster standard is in the works.

According to CNet’s coverage, the industry is split on whether it favors a maximum speed of 400 GBps or 1 Tbps. 400 Gbps is "technically and economically feasible," says chair of the High-Speed Ethernet Consensus group John D’Ambrosia. That speed is reportedly backed by networking hardware companies, while the 1 Tbps target is coveted by their customers. D’Ambrosia says hitting 1 Tbps wouldn’t be pretty and that it could require as many as 80 separate physical connections for a full-duplex implementation.

The faster standard will be for servers, of course. Good old Gigabit Ethernet seems firmly entrenched on desktops and notebooks, especially given the comparatively slow speed of most Internet connections.

GigE’s 1 Gbps transfer rate should be sufficient for most needs. However, it doesn’t look all that impressive next to the 802.11ac Wi-Fi standard. 802.11ac hardware is available now and enables transfer rates up to 1.3 Gbps under ideal conditions. That’s due to be topped up by WiGig, which promises 7 Gbps speeds over distances of 30-60 feet. Perhaps it’s time for PCs to start making the transition to 10-gigabit Ethernet. Gotta keep up with the Joneses.

Comments closed
    • albundy
    • 7 years ago

    ” Good old Gigabit Ethernet seems firmly entrenched on desktops and notebooks, especially given the comparatively slow speed of most Internet connections.”

    Good ol’ Google Fiber to the rescue. once it hopefully spreads like wildfire, cable will be nothing but a dusty old fart in the wind. but all this talk about huge bandwidth has really no meaning to joe schmo whose HDD or maybe even SSD can barely even use 1Gbps, let alone anything higher.

    ” Perhaps it’s time for PCs to start making the transition to 10-gigabit Ethernet”

    that would be a complete wast of speed. a better focus would be on upgrading the slowest components.

      • just brew it!
      • 7 years ago

      The NIC pretty much *is* the slowest component in a modern PC, not counting the ones that don’t need to be fast because there’s a human in the loop.

        • Krogoth
        • 7 years ago

        Actually the slowest component on a system is ODDs, HDDs and a few older SSDs. 😉

        There’s only a handful of SSDs on the market that can fully saturate a 10Gbps Ethernet connection.

          • just brew it!
          • 7 years ago

          Most current PCs don’t have 10Gbps NICs.

          • jensend
          • 7 years ago

          There is only one way your use of “handful” here can be justified:

          “I have in my hand an Intel 910 800GB PCIE SSD and I can’t carry anything else in that hand.”

          AFAIK no other drive will break 1GB/s even on sequential read.

    • Madman
    • 7 years ago

    Drop the cooper, and use optical interconnects everywhere! No problems with different grounds, no signal attenuation, no crosstalk, cables are thinner, it’s a massive win everywhere.

    And cables will get cheaper once used. That’s how market works.

    Or at least use some sort of a thunderbolt like active cable, that can be either cooper or optical.

      • Bensam123
      • 7 years ago

      Buy equipment to make your own fiber cables and talk to me again… And no, it’s highly unlikely the equipment for terminating your own fiber ends will get cheaper.

        • Krogoth
        • 7 years ago

        Outside of networking offices and datacenters. You don’t need to craft your own patch cabling. You can get standard lengths (2M, 10M, 25M, 33M, 50M) on the cheap if you know where to look.

          • hansmuff
          • 7 years ago

          Yeah I disagree strongly there; I wired my house with CAT5e. To do a proper job, you need wall plugs and such, so it isn’t like you can wire a house with pre-terminated cabling.

          • Bensam123
          • 7 years ago

          What hans said… I’ve wired quite a few houses while renovating them. I’m not buying a $3000 (which is cheap) thermal fuser splicer so I can wire someones home with fiber. I’m also not going to figure out every single run I do and write down the length to order them all online, then find out some are bad and have to go through a RMA process for a few dozen cables.

          I didn’t mention this, but running fiber through walls is a lot harder then copper due to bend radii. This isn’t the same as running a fiber cable along the ground, around your desk, and up to your switch.

      • sweatshopking
      • 7 years ago

      copper*

        • Madman
        • 7 years ago

        Damn, I misspelled it twice, and I actually knew how the word is written. I must have been drunk.

        Thank you!

      • just brew it!
      • 7 years ago

      Ethernet already deals with different grounds quite well, as it is differential and transformer coupled. In case you hadn’t noticed, standard UTP Ethernet cabling has no ground connection.

      Optical fiber attenuates too, unless you’re willing to pay for the really expensive stuff.

      That said, I agree with you – we should be going optical.

        • Madman
        • 7 years ago

        As far as I’ve heard, if you have 2 PCs in to different buildings connected with 100m of UTP cable, and each of the PCs is plugged in it’s own grounded power outlet, there is a great chance you might get a large spikes of electricity traveling over the wire.

          • just brew it!
          • 7 years ago

          Well you heard wrong.

          The only connections to ground on a typical Ethernet run are through the transformer windings and a 75 ohm resistor in series with a 1000 pF capacitor. The RC time constant on the capacitor/resistor combo is 75 nanoseconds, so it is effectively an open circuit until you get up into the hundreds of KHz (i.e. DC and power line frequency voltages are effectively blocked).

          I suppose if you’ve got shielded cables (note: most Ethernet cables are UNshielded, hence the ‘U’ in ‘UTP’) you could end up with some significant ground loop currents through the shield. If a significant difference in ground potential exists then shielded cables should have their shield connected to ground at only one end (which unfortunately will also render the shield much less effective).

          If you’re talking about induced surges from nearby lightning strikes that’s another matter entirely; I’m only addressing the case of different ground potentials.

            • Madman
            • 7 years ago

            Ok, that’s a bit more of electrical engineering that I can grasp, but I believe you.

            It’s funny how everyone at school thought STP is better than UTP, but in real life it’s still UTP that’s the king.

            • just brew it!
            • 7 years ago

            STP [u<]is[/u<] theoretically better from an EMI standpoint (both emissions and susceptibility). But UTP is cheaper, "good enough" for Ethernet in the vast majority of cases, and doesn't suffer from potential ground loops. Is it any wonder that nearly everyone still uses UTP?

            • Bensam123
            • 7 years ago

            Are you saying it’s better to run UTP outside in the ground then STP or ScTP? Because I heard and common sense would dictate otherwise. :l

      • superjawes
      • 7 years ago

      Barring any massive spike in copper prices, optical cables will always be more expensive as it is nearly impossible to splice (okay, not impossible, but you need special tools and a microscope to do it). And it can run into problems if you bend it, making routing more difficult.

      So you will continue to see enterprises use and adopt optics, but it is unlikely to see Ethernet replaced in consumer space.

        • Madman
        • 7 years ago

        Yeah, but they take no space, hence you can run them through those decorative planks between a wall and a floor. Or build in inside drywall.

        Then you can just multiplex stuff, and feed all the network devices for years to come through that one tiny cable.

          • superjawes
          • 7 years ago

          You can’t do that with ethernet? For way cheaper?

          I can get a 10 ft ethernet cable for $1.68 from Monoprice in 10 different colors, a 25 ft for $3.30, and a 100 ft cable for a whopping $11.17. Until optics can comete with those prices, ethernet will be the standard for years to come.

          And that’s ignoring the costs of getting fiber ports installed in your routers, computers, etc.

    • 0g1
    • 7 years ago

    Ethernet sucks. Stuck at 40gbit right now. Never reaches peak potential due to TCP/IP congestion problems.

    Infiniband FTW. FDR at 56gbit. Cheaper. IPoIB reaches peak potential bandwidth of the network due to congestion avoidance and reliability built into the network switch.

      • Krogoth
      • 7 years ago

      is the congestion problem at Level 3 (packets) or Level 1/2 (frames)?

        • 0g1
        • 7 years ago

        Frames and packets are the same thing.

        Packet level (3) TCP-IP (or any other protocol but this is the main one needed to be compatible with existing systems) is the only thing ethernet has. TCP is time based, send, hope to receive, if not in time period, slow down. If not received, re-send, no guarantee.

        Infiniband has a credit based protocol which is implemented in packets. But where it differs from ethernet is the actual Infiniband hardware switch uses an architecture which provides different levels of importance to packets and there are no dropped packets because the switch and adapter are working together, looking at the packets, and deciding when to send based on a bandwidth credit scheme and a priority lane scheme.

    • axeman
    • 7 years ago

    802.11ac hardware is available now and enables transfer rates up to 1.3 Gbps under ideal conditions.

    “Ideal conditions” don’t exist. In the real world, wireless continues to mostly suck. Maybe 802.11ac might change my mind, but wireless has failed to impress. The real world speeds are abysmal usually. I’ve yet to experience a wireless connection that can even match a 100mbit ethernet connection in actual speed, even if the link speed is reported at ~270mbps with 802.11n.

    Predition: even 1gbit ethernet will continue to be infinitely superior to 802.11ac

      • willmore
      • 7 years ago

      You either have some bad interference sources or you’re using crap wireless hardware. On only a 40MHz 802.11n single stream link, my routers bump into the backhaul before they run out of bandwidth on the RF side.

      On my 5GHz WAN, I’m running into the limits of the older hardware I have. I can’t seem to get more than 150Mb/s on that thing.

    • burntham77
    • 7 years ago

    Aren’t most desktops held back by drive speeds? Can even the fastest SSD saturate a 1Gbps Ethernet connection?

      • Elsoze
      • 7 years ago

      yes. 1Gbps = ~125MBps. Single SSD’s can sustain greater than 500MBps. More in raid configurations.

        • BobbinThreadbare
        • 7 years ago

        minus overhead too

      • ludi
      • 7 years ago

      This isn’t about desktops.

    • d0g_p00p
    • 7 years ago

    I’m all for 10GigE at the desktop, I am NOT however down for hi speed Wi-Fi just due to the transmitter power required to push that speed. I want the least amount of electromagnetic radiation being pumped though my head and body.

      • Spunjji
      • 7 years ago

      Time to move to Outer Mongolia, then! 😀

      • bcronce
      • 7 years ago

      Don’t worry, your light bulb puts out higher energy photons and MUCH MUCH more of them.

    • Bensam123
    • 7 years ago

    I don’t think gigabit is ‘firmly entrenched’ as much as it’s the default choice because there are no other choices to choose from. 10gigabit doesn’t have entry level equipment either for switches or nics, it’s just not there. Gigabit had entry level switches you could purchase when it started coming down in price, but 10gig has been around forever and we still have no entry level equipment

    I don’t know what he’s talking about in terms of 1TB requiring 80 physical links. Maybe with todays standards, with todays links, using todays equipment. The whole point of designing a future specification is so we can hit that goal.

    They shouldn’t be designing it around what we ‘need’ today too. That’s like the worst idea ever. Design for the future… always. Push it as far as you can go without having to design the entire computer around the damn thing. That’s what wireless has been doing and they went from 1Mbps to 1.27Gbps in the same time Gigabit has sat at the same link speed. We should already be at 1Tbps.

    I really have to question this… You know how much cheaper it would make bandwidth online if we had links going at 1Tbps?!!? Maybe this is exactly why we are still stuck at 1Gbps… It really makes me wonder (and should make you).

      • Anomymous Gerbil
      • 7 years ago

      Yep, these idiots discussing new standards just don’t have a clue. Give them a call.

        • Bensam123
        • 7 years ago

        I’m not questioning their intelligence as you imply by calling them ‘idiot’s, but rather their motivation and possibly motives.

      • bcronce
      • 7 years ago

      Going from 1Mb/s to 1Gb was relatively easy because you didn’t need much beyond some moderate electrical engineering. 40Gb and higher speeds are starting to run into some interesting physical issues. Electrons are to behave strangely and LEDs are getting pushed to their physical limits.

      A few years back, Intel said around 2014, they will start releasing integrated 10Gb nics in desktop chipsets. This may have been pushed off a bit because of 22nm issues, but not by much. Expect 10Gb in desktop computers in the next 3 years.

        • Bensam123
        • 7 years ago

        Yet we don’t have these issues over the air with all sorts of interference? This seems highly unlikely. Bandwidth doesn’t necessarily relate to how fast electrons move through a cable either… That has more to do with latency.

      • bcronce
      • 7 years ago

      “You know how much cheaper it would make bandwidth online if we had links going at 1Tbps?!!? Maybe this is exactly why we are still stuck at 1Gbps… It really makes me wonder (and should make you).”

      Blame incumbent ISPs.

      Internet backbone costs per unit of bandwidth is dropping 50% every year, yet my ISP increases me bill while giving mild bandwidth boosts with data-caps.

        • Bensam123
        • 7 years ago

        Yeah, I’m sure ISPs are to blame as well… But if you have links operating at 1Tbps at home it really makes you question your 1.5Mbps DSL connection you pay $20 a month for.

        • Krogoth
        • 7 years ago

        The problem isn’t the backbones.

        It has always been the last mile connection.

        USA has a ton of legacy infrastructure (old cables and POTS) and replacing all of that with fiber connections isn’t cheap or quick to do. Even if Telco companies went out of their way to do it. They still have to deal with local politics and NIMBY crowd.

      • Krogoth
      • 7 years ago

      Correction, 10Gbps Ethernet has been available for optical connections since ~2005. The copper version has only been available in more recent years. It has more physical limitations than 1Gbps Ethernet (short maximum lengths and more prone to EMI)

      The problem has backwards compatibility with existing Ethernet standards and trying to do it over copper (UTP). That’s why I believe we should be moving towards optical connections. Intel has the right idea with Light Peak.

      The lack of demand on mainstream end is preventing widespread 10Gbps Ethernet adoption. The mainstream prefers wireless Ethernet (less messy and more convenient). Despite the known security, EMI and performance issues.

        • Bensam123
        • 7 years ago

        You can’t say there is no demand when you don’t even offer a product within grasp of the people you’re targeting. I WOULD buy 10Gbps if switches cost $200 a pop and nics $75. I’m sure I’m not the only one either… That’s what I paid to adopt Gigabit before the rest of the world.

        10BaseT over Copper has been around since 2006, 10Gig in general since 2001. None of the issues you mentioned are mentioned in the wikipedia article.

        [url<]http://en.wikipedia.org/wiki/10-gigabit_Ethernet#10GBASE-T[/url<] It offers less latency and the same range as 1Gb over cat6a, which is pretty common. Cat 6 being even more common and having the same price as cat5e (it has more twists). No one 'needs' to run cat6 though and this technology would work over cat5 or cat5e in short runs. I don't believe there would be any problems with backwards compatibility if other standard makers can do it. USB does it, PCIe does it, SATA does it... Heck SATA is transitioning to pci-e and it's still backwards compatible! The entire ethernet standard is based around redundancy and checks/balances in order to negotiate a proper link. You can even plug a 10baset half duplex connection into a gigabit switch and it works fine (or it should). Compatibility is a non-issue. I don't understand how there would be 'security issues' present in 10Gb that wouldn't be present in every other form of ethernet available. Performance issues? When you can buy a thermal fusion splicer for less then $3000 to make your own cables, fiber will take off. People actually run ethernet through their home and they don't go online to buy every single cable for each run. This is actually still a business, where other businesses higher someone to come and terminate their fiber cables after running them. Fiber is great and everything ONLY when your switch is sitting next to your computer and all you need is a little patch piece. If they can do all of this through the air, they can do it through a cable. I'm actually more interested in why Wikipedia, which usually presents a lot of information on the pros and cons of a technology in addition to improvements does not indicate any of this about 10Gig or 100Gig. There are just a few foot notes, without sources... Compared to say PCIE or Wifi, which detail in depth what they changed between versions and pitfalls.

          • Krogoth
          • 7 years ago

          Again, there’s no major demand for 10Gbps Ethernet outside of businesses.

          The people who have a genuine need for it are prosumers of some kind. They are more than happy to play the premiums associated with 10Gpbs because time is $$$$.

          Mainstream market has no killer app that “needs” 10Gbps ethernet connection, 1Gbps is still more than sufficient for this group.

          You “may” have a need for it, but you are part of a tiny minority. There’s no economic incentive for networking guys to go after you. They aren’t going to try to undercut their more lucrative business customers by offering affordable 10Gbps solutions to a tiny, but vocal minority who don’t want to pay enterprise-level prices for 10Gbps Ethernet.

          CAT6+ is the only copper standard that supports 100M for 10Gbit connections. CAT6 can do it, but it is limited to ~25M cable lengths. CAT5e can only handle a little over 1Gbps Ethernet connection.

          I said that security and performance issues were associated with Wireless Ethernet a.k.a Wi-Fi not 10Gbps Ethernet. Mainstream market perfers Wi-Fi over wired Ethernet because it isn’t as messy and it is more far convenient, despite its known problems.

            • Bensam123
            • 7 years ago

            And how do you demonstrate there is no demand when there are no entry level products for consumers?

            How many people do you think bought gigabit when it first came out? Why do you think there was a market for that? Do you think that introducing 10gig consumer devices wouldn’t drop the prices in the enterprise sector and more people would buy it? Do you think that pulling 10gig down to consumer level wont increase speeds and R&D at enterprise level? Do you believe that file sizes haven’t increased almost exponentially since then? Average HD in 2001 was like 80GB… yet we had gigabit back then.

            I do appreciate you throwing around words like ‘vocal minority’ to try and set me apart, really I do, but I’m not the only one that would buy 10gig not just because I can use it, but because upgrading speed always helps. If it speeds transfers across my network from a few minutes down to a few seconds I’m all for it.

            Do you have any sources for your lengths and stats?

            I misunderstood that and even if ‘mainstream’ prefers wireless that doesn’t mean we can’t improve wired speeds. Fuck dude you can’t be so obtuse that you actually don’t want to be able to upgrade your computer parts and rather be cynical and jaded towards any sort of progress for some odd reason…?

            Like a lot of your posts about pushing technology always seem to center around ‘good enough’, ‘we don’t need this’, ‘What we have is fine’… You sound like a old fogy, only on a technology website for some odd reason.

            • just brew it!
            • 7 years ago

            [quote<]I do appreciate you throwing around words like 'vocal minority' to try and set me apart, really I do, but I'm not the only one that would buy 10gig not just because I can use it, but because upgrading speed always helps. If it speeds transfers across my network from a few minutes down to a few seconds I'm all for it.[/quote<] While I agree that there are situations where the extra speed would be useful, most home users simply do not routinely need to move 10s of GBs of data between PCs on their LAN, nor do they particularly care that it takes 5 minutes instead of 1 on those rare occasions when they do. That said, this is an enthusiast site. We're *all* effectively part of a "vocal minority"...

            • superjawes
            • 7 years ago

            [quote<]That said, this is an enthusiast site. We're *all* effectively part of a "vocal minority"...[/quote<] QFT

            • Bensam123
            • 7 years ago

            Do you believe that joe schmoe going to best buy to buy a wired switch knows or cares about what they need? Not everyone can quantify purchases based on their usage patterns (I would say that’s about as big as the vocal minority you’re talking about), but there are a lot of people that simply want the fastest of something. They can’t do this as there is no 10gig switches or nics sold at best buy, for good reason, because there are none within consumer price ranges.

            Personally I didn’t know 10gig has a latency reduction… that almost interests me more then the speed.

            • just brew it!
            • 7 years ago

            Gbit Ethernet already has round-trip latencies well below 1ms (on the order of 100us based on a couple of quick ping tests I just did on both my home and work networks).

            Just out of curiosity, what do you think the real-world benefit of reducing your round trip latency below 100us will be? Unless you’re running applications which are highly sensitive to microsecond-scale variations in latency (are you running a massive data center or HFT operation out of your house by any chance…?), the difference will be imperceptible.

            • Bensam123
            • 7 years ago

            Having my ping read 0 in games makes me happy.

    • Chrispy_
    • 7 years ago

    Desktops need something faster than gigabit badly;

    Part of the blame goes to companies like Autodesk and Adobe whose products generate files much bigger than they did when gigabit ports started becoming commonplace on the desktop.

    A decade ago I would have balked at a 200MB single image. Now I commonly see 4GB .psd files thanks to multiple layers, command history and undo data.

    The other part of the blame goes towards our incessant push for higher resolution cameras, screens, textures etc.

      • shiznit
      • 7 years ago

      Depending on how many users, at some point it would be cheaper to do VDI with FC connectivity to the SAN vs giving everyone’s desktop 10Gb ethernet. 100Mb ethernet is plenty for a thin client.

      This is referring to a big shop of course, at home you are probably better served by Thunderbolt/eSATA/SAS.

        • Chrispy_
        • 7 years ago

        Thin clients don’t work for content creation.

        Our workstation users are rendering, [i<]needing[/i<] hardware accellerated 2D and 3D, generally never happy with the power you can get from a discrete box. I might be out of touch, but I never saw a thin client solution that made good use of client end resources. They're good for spreadsheet monkeys, but the gorillas I work with need polygons!

          • shiznit
          • 7 years ago

          Don’t be so sure.

          You can already pass through a graphics card to a power user’s VM and you can have several in a server hosting your VDI VMs. Also, full GPU virutalization is almost ready and it will allow you to distribute GPU resources across multiple VMs so they share the same GPU just like they share the same CPU, look up Nvidia VGX.

      • indeego
      • 7 years ago

      So remove those layers. Man people love to complain about more features. 4GB .psd files are fine. If you don’t need something, find something that fits your needs.

    • Shambles
    • 7 years ago

    [quote<]However, it doesn't look all that impressive next to the 802.11ac Wi-Fi standard. 802.11ac hardware is available now and enables transfer rates up to 1.3 Gbps [b<]under ideal conditions[/b<].[/quote<] So under real world conditions what are we going to expect. 200Mbps? Just like Wireless-N it will over-promise and under deliver. Exponentially so the more dense your residential zoning. I am still, and will likely continue to favour a hard-lined connection for quite some time.

      • axeman
      • 7 years ago

      Upvoted, I said the same thing before seeing your older post. I only resort to wireless as a last resort, even using a USB to Ethernet adapter on my Wii.

        • swaaye
        • 7 years ago

        Wii has only 11mbps support I think. It’s so slow it’s a bottleneck even for internet access.

    • Peldor
    • 7 years ago

    [quote<] WiGig, which promises 7 GBps [/quote<] Well, not quite GBps.

    • maxxcool
    • 7 years ago

    Just give me 3gbs over wifi, that is all i ask, then i can boot a box without a hard drive over wifi with some reasonable latency…

      • Bensam123
      • 7 years ago

      Bandwidth isn’t the same as latency.

      • Beelzebubba9
      • 7 years ago

      Oh God – the idea of iSCSI over WiFi just made me wince. Disk IO protocols do not like dropped packets….

    • Krogoth
    • 7 years ago

    It is time to go completely optical.

    Copper is running out of room and it is becoming more expensive to use.

    Optical is not only faster and it still has headroom. It is practically immune to EMI and for you paranoid-types, it is very hard to splice/eavesdrop an optical connection.

    Optical Ethernet’s widespread use was held back by cable cost and cables being too fragile. The newer generation of optical cabling is more sturdy and cost per meter is close to copper wiring.

      • bcronce
      • 7 years ago

      Optical has only recently gotten cheap enough and flexible enough for ISPs to start rolling out fiber to the home.

      From what I’ve read, fiber used to cost closer to a few $ per foot, but now is only a few dimes per foot. Plus the old fiber would break from the slightest funny bend and was insanely hard to fix.

      This all happened in only the past 5-8 years.

      I’m sure your dream will come true soon(tm).

    • flip-mode
    • 7 years ago

    Why not just WiGi?

      • DancinJack
      • 7 years ago

      How would you pronounce that?

        • yogibbear
        • 7 years ago

        woo ha ha what what!

        • Bensam123
        • 7 years ago

        ouija

        • willyolio
        • 7 years ago

        oo eee oo a a ting tang walla walla bing bang

        • demani
        • 7 years ago

        Wigitty wigitty wack?

        • willmore
        • 7 years ago

        Like Will Smith. But you’d have to buy a new laptop to get WiGi with it.

      • krazyredboy
      • 7 years ago

      WiGi, WiGi, WiGi, WiGi, WiGi, yeah…WiGi, WiGi, WiGi, WiGi, WiGi, yeah…I’m a Sexy Network and I know it…

      Ok, so…it doesn’t quite, work like it did in my head…but I’m sure some of you will catch on.

    • sircharles32
    • 7 years ago

    Forget 10-gigabit on the desktop (too damn expensive right now). How about we first get motherboard manufacturers to start standardizing on two gigabit ports, with ganging capability. That in itself would be a huge boost in local file transfers (about 238 MB/s). With SSDs though, you’d ideally need about 5x the bandwidth of gigabit ethernet to keep pace with a SATA-600 SSD.

    That’s my feeling, as a home user.

      • shank15217
      • 7 years ago

      with 10gig copper card available I disagree, we need affordable 10 gig copper cards and switches.

      • Krogoth
      • 7 years ago

      10 Gigabit is in the same area that Gigabit was back a decade ago. It was enterprise-only along with the price tag.

      It will eventually come down in cost for NIC, wiring and networking equipment.

      • just brew it!
      • 7 years ago

      You need support at the switch level (Link Aggregation Control Protocol) to do ganging effectively. So in addition to getting the motherboard vendors to start putting 2 ports on the motherboards you’d need to convince the switch vendors to incorporate LACP into their consumer-level switches. I see a bit of a chicken-and-egg issue here.

      • Bensam123
      • 7 years ago

      Can be done with Intel nics. It’s built into their drivers.

      • shiznit
      • 7 years ago

      IP hash load balancing doesn’t work like that.

    • designerfx
    • 7 years ago

    400 GB/s? That’s it?

    *sigh*. This isn’t even close to enough for our network. This is a disappointment that they are not looking at 4TB/s by now. Switches and servers themselves are handling it with no problem, and have ports used up due to how ethernet is set up (multiple ports, etc). So much for a future-focused standard.

      • NeelyCam
      • 7 years ago

      So, you want 4TB/s when 1Tb/s is barely possible…?

        • designerfx
        • 7 years ago

        Maybe it’s because we’re not doing very well when it comes to advancements of ethernet? I do work for a large engineering/research company so we do process enormous amounts of data on a frequent basis. However, the fact that we’re not even looking at feasible 1Tb/s is indeed a problem.

      • Krogoth
      • 7 years ago

      Blame backwards compatibility and trying to make it work over copper.

      • JrdBeau
      • 7 years ago

      I’m curious what systems support 4TB/s. Is that for a single connection or an aggregation of a bunch of ports?

      Also, let’s be sure we’re getting our “B” and “b” correct.

        • chuckula
        • 7 years ago

        Two options:
        1. DesignerFX is in charge of all the data pumped out by the LHC at CERN.
        2. DesignerFX is full of it.

        I’ll let you decide which one you think is more likely.

          • bcronce
          • 7 years ago

          Infiniband + RDMA?

          If DesignerFX actually is in charge of data over at LHC, that I am curious to his disappointment(assuming it’s not sarcastic disappointment).

          A few ways to be disappointed is to have too high of expectations because of:
          1) Reasonable expectations
          2) Unreasonable expectations
          3) Sarcastic expectations

          If he had reasonable expectations, I would like to know what he expected higher speeds
          If he had unreasonable expectations, that’s just not being knowledgeable enough.
          Sarcastic expectations are a fun way of saying “I wish”

          From all the literature that I have read in the past 5+ year, 100Gb+ is a real pain because of physics, and I would expect someone at LHC to appreciate these issues.

            • JrdBeau
            • 7 years ago

            If I recall correctly, LHC’s grid network uses 10Gb links… but lots of them!

            Not sure even 12x Infiniband can hit those speeds.

            • designerfx
            • 7 years ago

            I don’t work at the location of the LHC, haha. That would be awesome and entertaining, but no. JrdBeau is correct in that it’s multiple connections – my point is that substantially higher speeds would simply reduce the requirement for multiple links.

        • designerfx
        • 7 years ago

        You are correct, it is aggregation. I’d like to see it from a single connection but alas, the point of my comment: we’re kinda behind on ethernet technology a bit. I do understand there are physical limitations at play here but I am not joking about how even 1TB/s is not sufficient in some instances already.

      • Anomymous Gerbil
      • 7 years ago

      So if even a massively-faster new standard isn’t fast enough for you, what do you use today?

      • bcronce
      • 7 years ago

      They have 1Tb+ links already, but it’s all custom hardware and costs a lot of money. What they want is *cheap* standardized 400Gb. They can always team the connections.

        • designerfx
        • 7 years ago

        Cheap? Cheap means it is probably done wrong. I’d like to see it done right. If it costs a lot of money, that’s not the issue. However, ethernet would make things a bit easier to deal with than FC and require a lot less specialization. Otherwise, what’s the point of ethernet at such speeds when FC is already handling it?

          • bcronce
          • 7 years ago

          My $0.50 integrated 1Gb NIC has more features and better reliability than the $100 1Gb nics from 10 years ago.

          If it get’s standardized and mass-produced, it will get “cheap”. Like cloths, just because you pay more for it, doesn’t make it automatically better.

      • 0g1
      • 7 years ago

      … your network? You don’t have switches that can handle it. If you know how the transfer worked … 10gbit serial means at least 1ghz transmission after say 256 QAM. You actually have to have a processor in your switch to convert that analogue into digital running at 1Ghz at least. Say you could get that to run at 5ghz, get 50gbit, have 4 lane transmission, get 200gbit. Still 20 times slower than 4tbit (I assume you don’t mean byte).

      You don’t have a server that can handle 4tbit/sec = 0.5tbyte/sec. Your pci-e 3.0 8x network card is only limited to 0.008tbyte/sec = 8gbyte/sec for a start. Say you could bypass that limitation and stream directly to system memory. What do you have … DDR3 2000? 2000x4bytesx4channel = 32gbyte/sec. Dude, still not 500gigs/sec. What if you could wire directly to your GPU memory (LOL @ wires connected to all those channels)? 6000mhz*48bytes=288gigs/sec. Nup, still not 500gigs per sec lol … fail.

    • chuckula
    • 7 years ago

    [quote<]IEEE taking input on next-gen Ethernet standard[/quote<] Here's what it looks like: 11010101010111010101010100111101010101000111101010101010 . . . 00101010100100001110101110010100100010100[b<]2[/b<] DOH!

      • way2strong
      • 7 years ago

      It was just a dream Bender; there’s no such thing as two.

      • bcronce
      • 7 years ago

      I thought it was funny :p

      TY for the joke

    • anotherengineer
    • 7 years ago

    Scotty…………….we need MORE POWER!!!!!!!!!!

      • Sargent Duck
      • 7 years ago

      LUDICROUS SPEED!

        • 5150
        • 7 years ago

        This thread has gone The Plaid.

          • DancinJack
          • 7 years ago

          You guys took Star Trek and turned it into Spaceballs. For shame.

            • superjawes
            • 7 years ago

            I find your lack of faith disturbing.

            • DancinJack
            • 7 years ago

            Trek is better than Star Wars and Spaceballs combined.

            • bitcat70
            • 7 years ago

            Of course it’s better. Star Wars and Spaceballs are so out of phase than when combined you get almost nothing.

    • sweatshopking
    • 7 years ago

    that’s pretty damn fast

    first

      • sonofsanta
      • 7 years ago

      Shame that “first” malarkey doesn’t mean much when the comments are re-ordered as they’re added, eh?

      • superjawes
      • 7 years ago

      Looks like being “pretty damn fast” is a problem for you.

        • yogibbear
        • 7 years ago

        He won’t admit it, but we all know it’s true. Even his touch screen TV has longer latency than his pants.

          • sweatshopking
          • 7 years ago

          that’s true. it doesn’t take me long till i’m READY TO GET TO BUSINESS. BUT ONCE WORK STARTS, I CAN GO ALL DAY.

Pin It on Pinterest

Share This