back to article Nokia, ARM, twisting Intel bid to reinvent the TCP/IP stack for a 5G era

Nokia and ARM are at the heart of a bid to revamp the ageing TCP/IP stack to make it better suited to networks that need to operate at very high speed and/or low latency. Among the plethora of industry alliances at the intersection of telecoms, IT and IP, this looks like one with a genuine contribution to make to the evolution …

  1. James 51

    If Nokia can get some of their patents in the standard the FRAND fees might be a nice little earner for them, particuarly if every I.O.T. device has to give them a penny or two.

    1. Harry Kiri

      Good point, there's more than just TCP/UDP in the IP world. How about XTP which is free and open. I coded one in a BSD stack in 1999, oddly nuff to optimise comms for a Telecoms company as TCP didnt cut high throughput, high/low latency etc etc. And then I did it again in a linux stack for a SatCom operator.

      Funny thing is TCP has lasted incredibly well, all things considered.

      1. HmmmYes

        TCP is a compromise. That's why its lasted, that's why people keep trying to re-write it.

        1. Anonymous Coward
          Anonymous Coward

          Nope its not. Its designed on the premise that lost packets are caused by congestion on layer 1, so throttles back to not pour petrol on the fire. That assumption doesnt hold with an RF physical layer as you lose packets due to drop out. And if you care about long latencies on acks slowing you down you can open the tx windows massively.

          So it was built to solve a different problem, that it works so well says as much for the reliability of RF comms...

          1. HmmmYes

            Preventative cyclic re-transmission is your friend for long latencies.

            Differeent protocols are need for a long bit of wire, and 100miles of air.

            1. Anonymous Coward
              Anonymous Coward

              "Differeent protocols are need for a long bit of wire, and 100miles of air."

              "Differeent protocols are need for a long bit of wire, and 100miles of air."

              Maybe.

              How familiar are you with the low level innards of the world of DSL? I used to be a bit, not so much these days, so please feel free to correct any errors below.

              The lower layers in the DSL stack have error detection, and in some circumstances, error avoidance. Error avoidance (interleaving) introduces extra latency on the millisecond scale, maybe up to a few tens of ms.

              That obviously slows down error recovery in the higher layers if something does go wrong, so lower layer errors which become visible higher up are a Bad Thing.

              So much of a Bad Thing are they that the DSL broadband industry has worked to define and adopt (and is now deploying) a scheme where many/most lower layer errors are now detected *and where possible recovered* in the lower layers (by lower-layer retries). This is the magic world of G.INP (Impulse Noise Protection).

              Now obviously the wireless world of networking and the wired world of DSL have nothing in common and nothing to learn from each other. Because of (e.g.) the difference in latencies.

              Or do they?

      2. Suricou Raven

        No, there's just TCP and UDP. Other transport protocols are an option only if you control the network end-to-end, because most devices on the internet these days are behind PAT, and PAT routers are only programmed for TCP and UDP. That's why TCP lasts: When the network is no longer dumb, introducing new technology is a nightmare.

  2. akeane

    Why did the BBC...

    ... reinvent PF Ring and DPDK?

    1. Warm Braw

      Re: Why did the BBC...

      I think that was covered here. [Hint: they didn't]

  3. Warm Braw

    "optimized for ... massively varied use cases"

    I take it that means "less-optimized for specific use cases"?

  4. Chris Miller

    What about IPv6?

    Much better suited to mobile devices (mobile IPv4 is a monumental kludge), and the only way IoT, bringing an order of magnitude more connected devices, can reasonably work (whether or not you think IoT is a good idea, the industry sees this as their next big growth opportunity).

    1. Mage Silver badge

      Re: What about IPv6?

      No, IPv6 mainly addresses the addressing issues. It does have some other "fixes", but still inherently not good for Wireless. If it was, then Satellite systems would use it natively on the links (nearly 90,000km each way, ground station to user.), they don't.

      Mobile / Wireless needs a protocol designed for massively variable speed, packet loss and latency, which can vary from excellent to really bad in minutes for a given link. Also the number of user connections per sector can vary from 1 (20Mbps) to 20+ (0.12Mbps) during a transaction without the addition of weather, movement, interference (even from other cells) or whatever affecting the speed, latency and packet loss. IPv6 doesn't make a huge improvement over IPv4 for that scenario. Also Mobile has the philosophy that individual users get all of the bandwidth available, rather than throttling back power (reducing inter cell interference and increasing battery life) when 1 or 2 users compared to 5 users in a sector. It's pretty garbage anyway for more than 10+ serious users of data (c.f. 3G where about 100 native voice calls are possible).

      4G is really really inefficient for voice calls compared to 2G and 3G native voice frames as it's carried as VOIP using TCP/IP to setup and control, then UDP streams. Hence most operators with 4G & 3G make your voice calls use 3G or even 2G!

      1. Anonymous Coward
        Anonymous Coward

        Re: What about IPv6?

        "... make your voice calls use 3G or even 2G!"

        and your point is?

        Why on earth would I care how a voice call is carried? They can route it via copper cables as far as I could care, as long as it works and is stable.

        1. James 51

          Re: What about IPv6?

          Voice HD or what ever they are calling it now. If you've got compatible handsets over a compatible network it does make a big difference. When a friend bought a phone that was I was able to tell the difference immediately.

          1. Mage Silver badge

            Re: What about IPv6?

            3G has "voice HD". It has a choice of codecs. 4G can't use some of the 3G and 2G codecs as they are less than optimal for TCP/IP UDP type networks. VOIP needs codecs optimised for TCP/IP unless you know the latency and packet loss is very low, even then the native "wireless" codecs have wasteful frame sizes for IP traffic

        2. Mage Silver badge

          Re: Voice on 4G

          You'd care if other people's voice calls was only carried on 4G. The data speeds might be 1/4 for you.

      2. Chris Miller

        @Mage

        It isn't speed or latency that's the issue (though that could certainly be a problem too), it's Mobile IP - the need to maintain a static IP address while moving from mast to mast and (potentially) roaming between different subnets. Understandably (since it was decades before the first mobile data call), the problem was never considered in the design of IPv4, IPv6 was designed from the ground up to cope with mobile devices.

        1. Mage Silver badge

          Re: IPv6 Mobile

          " IPv6 was designed from the ground up to cope with mobile devices."

          Err.. citations?

          I've never ever seen that claimed before. Other than the addressing size and giving the same IMEI the same v6 IP every time. (Which might be a bad idea for privacy!)

          "the need to maintain a static IP address while moving from mast to mast and (potentially) roaming between different subnets."

          That's a completely separate issue. It's still not very reliable. Switching between 3G,4G and WiFi without losing a session is really hard. It's possible and IPv6 is nothing to do with it. Switching between mobile 4G sites with IPv4 works fine, was testing it in 2007.

        2. Roland6 Silver badge

          Re: @Mage

          "IPv6 was designed from the ground up to cope with mobile devices."

          Err b*ll*cks!

          The initial design work for IPv6 began in the late 1980's and was largely completed in 1998, ie. prior to any real consideration of mobile device networking needs. Don't remember mobile being mentioned in any meaningful way in any of the drafts and minutes that came across my desk.

          1. HmmmYes

            Re: @Mage

            No, IPv6 was designed by Marc Zuckerberg and Donald Trump.

            To be honest, as far as any of the IP protocols go, design is probably the wrong word. Maybe the first iteration?

    2. Roland6 Silver badge

      Re: What about IPv6?

      IPv6 does zero, to fundamentally change the "creaking TCP/IP" protocol stack. In fact given the significant enlargement of packet headers it probably makes matters worse. Advanced protocols such as XTP do radically change things - XTP is a single lightweight protocol that completely replaces the TCP/UPD and IP protocols and was designed to be used dynamic ad-hoc networks.

  5. Mage Silver badge
    Boffin

    Misleading.

    " optimized for the massively varied use cases of the next mobile generation, for cloud services, and for virtualization and software-defined networking (SDN)."

    Only wireless is different from point of view of TCP/IP stack and protocol compared to the others.

    "Cloud Services" = Remotely connected servers, nothing new here.

    "Virtualization" is irrelevant.

    " software-defined networking (SDN)" isn't really about TCP/IP at all. It's infrastructure management.

    Radio unlike cable*, fibre and in house networking suffers from variable packet loss (other fixed cable/optical links have almost none to zero) which can be high. Wireless has unpredictable variable latency and speed too. The other technologies at the link level are pretty much fixed latency and speed. So TCP/IP is really poor for wireless, especially outdoor wireless like mobile. The ultimate "wireless" is satellite. It doesn't bother with TCP/IP at all. Best to imagine each end of two way link is a pair of proxy servers with a special protocol between them. That's why a VPN is garbage performance on Satellite unless the satellite ISP has your VPN endpoint in your modem and recreating the VPN at their Groundstation.

    This is why 4G stuck with TCP/IP, though really a poor solution for mobile, as without being more clever than the Satellite modem folks, lots would be broken.

    So this is really ONLY about mobile and how to have not TCP/IP over the wireless transparently to all existing traffic. Look at IP V6. Is the end user or average business going to change to an alternative to TCP/IP?

    Anyone today can do their own design of TCP/IP implementation on their OS (virtualised, cloudy or not) as long as to the external network port that it meets the spec!

    [* DSL, VDSL etc and other schemes over Cat 3 (phone wires) is more like Radio than cable or fibre. Unlike actual outdoor Mobile, the latency and link speed is constant for a given pair of wires, but the packet loss / interference issues can be like WiFi or Mobile. A powerline Ethernet adaptor or electric fence can disrupt DSL]

  6. jonfr

    What about 2G, 3G, 4G?

    They haven't even finished building 3G network in many places and some areas are not even close to have 4G connection and they are going directly for 5G already.

    This is not going to end well I say.

    1. Mage Silver badge

      Re: What about 2G, 3G, 4G?

      No-one is "going for 5G" already. It's still an ill defined concept, possibly more about integrating infrastructure. There is no 5G standard yet, nor any 5G base stations or software or hardware.

    2. James 51

      Re: What about 2G, 3G, 4G?

      If I remember my el reg articles correctly 4G was a term coined by the marketing rather than the engineering department. 5G maybe the same. I take your point about network coverage though. Will handsets have to support three generations of wireless standards and their various flavours around the world? Could end up pushing the price of phones up.

      1. Fatman
        Joke

        Re: What about 2G, 3G, 4G?

        <quote>If I remember my el reg articles correctly 4G was a term coined by the marketing bullshit artists rather than the engineering department.</quote>

        FTFY!!!

      2. jonfr

        Re: What about 2G, 3G, 4G?

        "If I remember my el reg articles correctly 4G was a term coined by the marketing rather than the engineering department. 5G maybe the same. I take your point about network coverage though. Will handsets have to support three generations of wireless standards and their various flavours around the world? Could end up pushing the price of phones up."

        They also have to support wide range of frequencies. My newest mobile phone supports.

        GSM 850/900/1800/1900

        WCDMA 850/900/1900/2100

        LTE (4G) 1(2100), 2(1900), 3(1800), 4(1700/2100), 5(850), 7(2600), 8(900), 12(700), 17(700), 20(800), 28(700), 38(2600), 40(2300)

        What is missing is LTE 450 that is being tested and developed. It is going to be used in the Nordic countries (Iceland, Norway, Denmark, Sweden, Finland + Greenland and Faroe Islands).

        All this spectrum is big drain on the battery due to scanning. It also has NFC (13.56Mhz). Bluetooth (2.4Ghz) and WLAN 2.4Ghz and 5Ghz. This is a lot of antennas (some might be shared) for one phone and battery drain if everything is turned on. I try to keep most of this turned off.

        I got Sony Xperia Z5.

    3. Tom 7

      Re: What about 2G, 3G, 4G?

      I only get 2G half way up the lane, or sometimes in the bathroom but only if the header tank is half empty.

  7. Boothy

    @ James 51

    Yup, Technically 4G is just a tweaked 3G, and not actually a new generation.

    1. Mage Silver badge

      3G vs 4G

      Main differences:

      No native voice on 4G. 3G and 2G voice is native, early data modems simply used the voice modulation, hence 245kbps on 3G and 14.4kbps on GSM. HSPA on 3G adds non-voice modulation modes to the CDMA hopping carrier to increase data rate, depending on link S/N. GSM EDGE is also a trick using different modulation to original GSM to get 200Kbps + There is a version of GSM (not used) for data only called ERMES+ that can do 2.4Mbps and native IP only (no voice). Superior to 3G HSPA as there is no cell breathing and speed is consistent. But bad for marketing as can't compete on peak speed.

      3G has no native IP, IP is native on 4G

      3G is a wider band version of CDMA-1, typically 5MHz CDMA coded channels. 4G (on downlink at least) uses COFDM or even COFDMA (loads of separate carriers, in 1MHz, 2MMHz, 5MHz, 10MHz or 20MHz channels. GSM uses a 0.2MHz carrier. CDMA 1.25MHz, or 3G 5MHz, is a single carrier frequency hopping in a known pseudo random fashion. Each link is different key. If all keys are used up, then the data is multiplexed, voice and data are different for 3G and HSPA version of 3G.)

      Flash-OFDM, Wimax and LTE are 4G systems.

      WiFi can be CDMA or COFDM depending on version.

      There are real 4G specs, though the important ones for 4G mobile are all LTE, which is certainly NOT at tweaked 3G, it's as different as 3G is from GSM (2G). The 3G is however based on USA 2G (CDMA-1), I guess Qualcomm wanted more money.

  8. HmmmYes

    Why in fscksake would you get Intel involved in writing any software?

    Seriously. I've stuff that uses Intel software - ether Intel produced or contracted by Intel. Its a fscking nightmare. And Intel abandon stuff after a couple of years.

    Sure, spec out a new protocol to run alongside TCP. Just don't call it TCP.

    1. Anonymous Coward
      Anonymous Coward

      I've stuff that uses Intel software - ether Intel produced or contracted by Intel. Its a fscking nightmare.

      Trillium, perhaps? Nightmare indeed. The only commercial software I've used where patches, supplied in source form, didn't even compile, due to syntax errors.

      1. HmmmYes

        Trillium stuff is Intel state of the art, fully supported, laser-like focus on customer, etc etc.

        No, Im talking about some software for some god-awful useless chunk of silicon that Intel had decided was going to be the second coming - razzmatazz, millions spent of marketing, big fanfare - dancing elephants.

        6 months later, after it flopped, and the errata list exceeded War + Peace, and all the marketing had moved onto the next 'Big Thing! (c)' they flipped me some dire code in a feeble attempt to wash there hands of it all.

  9. Christian Berger

    Please don't

    If they already start with buzzword bingo the result is likely to be just a failure like HTTP/2.

    The whole idea about Internet protocols is that they are easy to implement by anybody. If you don't have a diverse range of implementations you end up with problems like Heartbeat or NTP vulnerabilities. To have a diverse range of implementations you need to have very simple protocols. The simpler the protocol, the better.

    "Optimizing" a protocol for any special usecase, i.e. mobile telephone networks, only means increasing the complexity while increasing the barrier for competitors to enter the "market". In the end you'll have some very few implementations with one or two which are good, while the rest is to bad to be used.

  10. Big Chief Running Bare
    Alert

    TCP/IP means IP plus transport, transport is the main problem

    Most of the criticism should be levelled at TCP and its ability to adapt to the network circumstances. Not sure whether IP is the problem, whether IPv4 or IPv6. Hosts want ubiquitous access to the internet, and IP networks bring that, especially in the nat-free world of IPv6. (Also doing away with intermediate fragmentation and mtu challenges is another unsung joy of IPv6, but I digress). So the main problem is the transport layer, and I think that is what the article is telling us.

    Now my other point is that all that is "payload" consideration in mobile networks. IP also exists as an underlay in mobile (between Base station and Core, in case of LTE). This, with the 3GPP protocol GTP over UDP over IP, provides the mobility tunnels that create the mobile bearer. Now it would be easier to design in a replacement to this layer (as opposed to changing the entire internet), as it is standardised by one standard body across all mobile operators, and remains transparent from the mobile users payload. Currently to open up mobile access to non-3GPP access networks, like WLAN and femto cells, another IP layer of IPsec tends to be required, with the GTP held within. This strata is ripe to be optimised! The desire is for 5G to be a unification of access networks, so an optimised layer providing security, mobility and universal connectivity beneath the internet payload would be the goal.

  11. Dinsdale247

    FreeBSD?

    I'm wondering what part of this is FreeBSD TCP/IP? The high performance userland network stack in FreeBSD is called netmap and it's a stand alone project:

    http://info.iet.unipi.it/~luigi/netmap/

    I don't see any reference to netmap anywhere (yet).

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like