Wednesday 7 August 2013

NBN: Turnbull's Quintuple-whammy for G.fast

Abstract: G.fast is "bleeding edge" technology, or "vapourware". It will need 20-30 times as many nodes, will need 10-20% more Fibre laid than FTTP, will need two-pairs to get 1Gbps (spares not currently available), the electronics will be 5-10 times more expensive than VDSL2, will need extensive remediation and rewiring of the copper network and will consume ungodly amounts of power doing this. The same applies to short-range wireless "picocells". They've been tried and aren't economic.

Turnbull is either technically illiterate or informed and misdirecting on a monumental scale. Which, given he has written that NBN Co is a purely commercial concern, raises serious questions about his statements under the CCC/ACL.


For around 15 years, 1 gigabit/sec over copper for 100m has been available, it's not new, nor difficult. My guess is there are 2-5 billion ports currently in-use. It's called Gigabit Ethernet, 1000BASE-T, and is the default chipset in every PC, laptop and average computing device. There are aleady 10Gbps copper links around, like Infiniband and the Apple/Intel "Thunderbolt". They either use very high-spec cables or are limited to about 2m: even in a machine room, 10Gbps Fibre ethernet is way cheaper, easier and more reliable.

The problem for Turnbull is that, as he says, "you wouldn't start from here". Real Gigabit over copper is symmetrical and starts with higher-spec cable (Cat-5a or Cat-6, Cat = "category"). The whole problem is that the telephone cabling we have does exactly what it was specified for in 1925: carries 4kHz very well. The 25,000 times faster 100MHz needed for "G.fast" is a very, very long stretch indeed.

The only way to get cheap, reliable 1Gbps over Copper on the last 100m is to run new Cables, as is done in every commercial LAN. But that's as much, or more, than Fibre - because the Copper is necessarily thicker and heavier than Fibre.

Turnbull need to answer a fundamental question:
Why is he so intent on protecting an asset, the Copper Customer Access Network, that doesn't belong to either the Government or NBN Co, but to an entity the Coalition valued so poorly, that it was sold, when he was a Minister, at knock-down prices in 2006: Telstra.

G.fast - "New Technology"

Turnbull doesn't use the correct technical vernacular for this untried, untested,  pre-production technology: Bleeding Edge. Technologists with more than 5 years experience are universally wary of big claims by any technology vendor. Most fail to perform in practice, only a small proportion of announced "revolutionary" products actually get to market, and then only a very few scale-up into volume production (like 1Gbps Ethernet on twisted-pair and 10/40/100Gbps Ethernet on Fibre.)

To run G.fast nodes in Australian conditions, the following apply:

  • A 100m limit means cables can't cross roads.
    • Nodes can only serve one side of the street.
  • The NBN is going to roll-out 148,000 road-kilometers past around 12M premises.
    • That's 12.5m per premise on two sides of the street,
    • or 25m/premise down one side of the street.
    • G.fast nodes will average, at most, 8-ports per node, with nothing left for expansion.
  • Long-blocks, long-driveways and battle-axe blocks allow only 1 or 2-ports per node.
    • The average count of ports per node will hence be much lower than 8.
  • Would any rational engineering project reserved G.fast for Flats & Apartments (MDU's), as in "Fibre to the Basement?"
    • NO. When 1Gbps ethernet is readily available, under 1/10th the cost of G.fast and only needs a few short runs of Cat-5/6 cable installed, a commodity these days, there is every reason to not use G.fast, or even VDSL2.

Turnbull fails to disclose the realities of his "New Technology" - it doesn't scale-up and worse:
  • At a minimum, Turnbull is going to need 20-times more nodes for G.fast.
    • That means digging up every single Nature strip to install both Fibre to the Nodes and Electricity to power those Nodes. This ensures colossal disturbance of every neighbourhood.
      • The cost of civil works for nodes doesn't scale with size, but stays roughly constant.
      • For an 8- or 12-port node, it will cost around 90-95% as much to install as a 200- to 400-port Node, or "RIM", CMUX or Mini-Mux as Telstra calls them.
  • Paul Budde, a globally respected Telecomms Analyst, reports that Scandinavian Telcos, even with VDSL2, are finding they need to run more Fibre in an Fibre to the Node Network than in an Fibre to the Premises Network. These are exactly the "only experts that matter" according to Turnbull, people in Telcos doing it "for real".
    • The reason is simple. The Fibre for Nodes has to be run down the street, past the houses to the node, then the copper comes back to the house.
      • The nett result is that G.fast at 100m down each side of the road, will require 10%-20% more Fibre to be laid than direct Fibre to the Premises.
  • As carrier frequency rises, so does the need for "remediation".
    • ADSL 1 at 1.1 MHz needed little work to allow 1.5Mbps at 4,000-5,000m.
    • ADSL2 at 2.2 MHz needed enough remedial work to get 12Mbps at 1500m that Telstra in 2005 noted it as their second biggest
    • VDSL2 at 17 MHz and 30 MHz over 800m will need extensive line testing and remediation to reliably achieve even the lowest denominator, 25Mbps.
    • G.fast at 100 Mhz over 100m won't tolerate any bad joints, any patching nor any degraded copper or insulation. "Water ingress" will likely stop the service stone-dead: an effective short-circuit.
    • As DSL speeds increase, the need to "remediate" (fix or replace) the Copper increases, a G.fast solution will need a massive amount of copper replaced, negating its cost advantage over direct Fibre.
  • G.fast only achieves 1Gbps is the following is done:
    • "bonded pairs": two sets of wires are needed, each giving
    • Virtual or "phantom" pair: an old telephony trick reborn. A third pair of wires is made
    • Vectoring: very fancy, very complex "noise cancellation". In electronics, fast + complex  equals really, really expensive.
    • For subscribers to get the advertised 1Gbps, they will need 2 pairs to their premises. The Customer Access Network simply doesn't have 100% spare lines installed waiting to be used.
    • As soon as new copper has to be laid, it's cheaper to run Fibre to the Premises.
  • The electronics for the G.fast lines, the line card, will be very complex to do the many complex calculations required, on the order of what is done in a high-def CAT scan, it also has to run very fast to handle the 100Mhz line frequency and they will not be cheap, high-volume production chips.
    • The electronics, line-cards, for G.fast will be 5-10 times more expensive than even VDSL2.
  • There's an iron-cald law of computing: the more calculations you want, the more power you need, the faster you want to do them, power rises quadratically (why we hit the CPU "heat-wall in 2003 at around 3Ghz).
    • These fast, complex, expensive and fragile chips will also consume 10-100 times more power than VDSL2.
    • The operational cost of powering these 1.25 million nodes will really be "another BIG power station." [Update: 9M lines ÷ 8 ports = 1.25 M nodes. @ 1+kWatt each = 1.5 GigaWatts]
    • These chips will run hot. The normal air-cooling in street-side cabinets will be insufficient - just like the fans on laptops and PC's go into overdrive on hot days, but eventually either turn-off the computer or slow it down to cope with the heat.
      • Either all these nodes will have to contain cooling systems for days over 35°C, or
      • they will throttle-back speed on "hot days", even shutting down altogether.
      • For every 5°C increase in "junction" operating temperature, component life is halved. These devices will need constant replacement in our environment.
  • Every time you install a piece of "street furniture", you are also committing to removing it and restoring the built environment.
    • The cost of the removal and disposal of the G.fast nodes has to be counted in their cost, not just "installation".
With all these negatives, one node per 8-houses, down each side of the street and disturbance everywhere for Fibre and Power, I can't see any sane Council approving their construction.

Hiding them "in-ground", in Pits, is a really, really bad idea. Pits will fill with water and they don't have good ventilation required for cooling. 240V + water + expensive electronics is a very bad combination.

The "gotcha" for G.fast is that it will interfere with FM radio - it works in the same frequency band, and those 100m twisted-pairs will make fantastic transmitting aerials. Interference with radios will be a constant and unavoidable outcome of the hare-brained technology.

All these arguments, plus others, apply to wireless "pico-cells".  "We can 'just' put up a wireless access point on every second power pole" is rubbish. A great idea until you try to do it: already has been tried by billionaire Paul Allen in California and Colorado (Ricochet) around 2000 and it failed miserably. The economics of fixed-line Fibre has continued since then to draw ahead of Wireless and Wireless-over-Phone-lines as DSL should be called.

Summary:

G.fast is a very poor technology, even if you own the Copper and want to protect that asset, which NBN Co does not.

If G.fast does turn out to be "30% cheaper to install than running Fibre" in densely-populated Europe or American cities, it won't be so in Australia. Everywhere here, Nodes have to cope with summer temps that will fry the electronics or required every node to be cooled, trebling the initial cost, doubling the running costs and decreasing the reliability of the service.

It's not even a good idea for Flats and Apartments: if you want cheap, durable and "cost-effective" 1Gpbs copper connections, run 1000BASE-T and put in new Cat-5/6 cable!
Fibre-to-the-Basement is a colossal waste of money.

At the very least, 20-times more nodes are needed than VDSL2, with electronics costing 5 times as much and 2-5 times as much 'remediation' of the existing copper and powering them will require "a BIG "power-station".

If a VDSL2 FTTN will cost $8 billion, then 1.25 million G.fast nodes FTTN will cost upwards of $50 billion.
If Turnbull isn't proposing to roll-out G.fast everywhere, then what does he mean by a National Broadband Network? Selective and highly-subsidised for the favoured few?

If Turnbull doesn't know this already, he's uninformed and incompetent.
If he does know this, then he's deliberately withholding material information when discussing a new commercial venture.

11 comments:

  1. Hey Steve.

    The Paragraph starting:

    As carrier frequency rises, so does the need for "remediation".

    You've Got Mbps after ADSL and VDSL. I think this is supposed to be MHz?

    Interesting write up otherwise. I don't think Turnbull would be stupid enough to try G.Fast everywhere to get 1Gbps....but then again.... :P

    ReplyDelete
  2. "That means digging up every single Nature strip to install both Fibre to the Nodes and Electricity to power those Nodes. This ensures colossal disturbance of every neighbourhood."

    G.Fast nodes are reverse powered from the end users premise over the existing copper pair. Thats a core concept of the technology.

    G.Fast nodes will *have* to be low power and waterproof due to the limitations of reverse power, and the nodes that exist already fit that requirement.

    ReplyDelete
    Replies
    1. ljm,

      The nature strips will have a node every 8 houses and mostly _less_.
      1.25 million of them will NOT be a good look.
      Even just laying the Fibre every 8 houses on both sides of the street _will_ mean a lot of disturbance and a lot of digging things up. And a LOT of heavy machinery to form slabs, install pits/junctions, hoist on heavy (= secure) cabinets and techs scurring around doing the cabling, testing and remediation for _days_ complete with pulling new cable and rejointing. There will be a mess in front of most houses.

      Not to mention, 'remediation' and cut-over on this scale guarantees service outages for everyone 'passed'. And what fault rate would you predict? The rejointing and recabling won't go without error. Touching 9M services _will_ create problems..

      My reading was they _may_ be reverse powered.
      Which is a far worse situation - you've just lumbered _the consumer_ with paying domestic/retail electricity prices to power the TELCO's box.

      It's not just the line card that has to be powered, it's the backplane, uplink and cooling system. I thought 30-75W budget for the line card and the same again for backplane, vectoring and "phantom" ccts (remember to get 1Gbps, need _two_ pairs, bonded).

      Then the same again for the cooling system. Can't just use fans and free air in our summer - because this gear will run HOT like a high-end Graphics Card.

      The highest spec of "power over signal" is PoE, which, IIRC, runs at up to 17V and 750mA. Can deliver 10-12W over a _limited_ distance, NOT over 100m.

      AWG 26 copper with low-voltage insulation is designed to carry 20mA at 50V = 1Watt.
      Admittedly over 4-5,000m. Would you run 40 times the juice, 8A, over 100m?
      They'd be _hot_ and the joints would fail.

      Are you actually suggesting they could run Nodes off the same 1Watt used for Telephones? No... So just how much power?

      I think from the compute task required, they won't be able to power a FULL system with less than 200W/port, including summer cooling.
      What happens when you have a node with only 2 or 4-ports active?

      Where do they get the power from for the backplane, uplink and fans/cooling?

      This won't be cheap for the consumer - either the "Power over Phoneline" device or paying for the electricity.

      Delete
    2. "Which is a far worse situation - you've just lumbered _the consumer_ with paying domestic/retail electricity prices to power the TELCO's box."

      Its not considered particularly egregious now when NBNco require the NTD to be home powered. They could have run an external power feed for those and installed them externally but it doesn't make sense. Especially considering...

      "Are you actually suggesting they could run Nodes off the same 1Watt used for Telephones? No... So just how much power?"

      Thats not only what im suggesting, its the reality. Read up on the technology. The reverse powered spec it has to comply with

      http://www.etsi.org/deliver/etsi_tr/102600_102699/102629/02.01.02_60/tr_102629v020102p.pdf

      Has a goal of 1W.

      And the nodes used in trials and prototypes by Alcatel etc. are reverse powered. There are even commercially available reverse powered nodes, although they dont currently implement G.Fast.

      Its a core part of the G.Fast spec that it has to comply with 1W reverse power, and nodes have to work if only 1 person is connected.

      With the capacity constraint of 1W on the existing line the power burden on the consumer is minimal.

      The nodes themselves are designed to be installed in existing pits

      http://www.belllabs.be/opendays/archive/2012/jochen_maes.pdf#page=7

      Slab's etc are not required and there would be no more disturbance than installing a fibre multiport in the pit.

      The potential saving here is substantial. NBNco themselves cost the customer connect at $1100. So g.fast needs to cost less than that for it to be feasable. On top of that there is the external benefit of avoiding up to 9 million days off work, as G.Fast is designed to be a self install technology.

      Thats obviously an upper estimate, some houses will require professional installs. But on the other hand NBNco themselves estimate about 30% of houses recquire more than one day for installs.

      G.fast is a new technology, and I would certainly bet on PON iterations as a more stable upgrade path.

      But misrepresenting the technology doesn't help your argument.

      Delete
    3. Whoever you are, I don't buy your argument. THERE IS NO PRODUCT YET.

      I've seen so much "vapourware" and ridiculous announcements of "product break-throughs" I've lost count. Vendor hype is rampant in this industry (not that we're alone).

      Until I see working, production equipment, in our conditions, I won't believe your claim of 1W per line. I know the sort of processing power you need to do the computing they want, it's around the same as a high-end Graphics Card. Lots of memory, lots of power, definitely not 1Watt.

      I refer you to Turnbull's argument: disbelieve any claim not made by people actually _using_ the equipment.

      It's not for sale, it's not in production, it's vendor hype until it is.
      I won't debate hypothetical specifications with you. COme back in 201 when there's a tested, proven product on the market.

      I note you didn't choose to argue we won't need 1.25 MILLION of these nodes.

      Delete
  3. All the other issues justify not installing this rubbish but I thought I’d point out:

    You also have the issue of the 70MHz band being used for critical services such as Emergency service and mission critical SCADA. Any system that goes much beyond ADSL2+ is starting to eat into that could be used for other services.

    Another thing:
    If Telstra does not currently have cooling on their top hats then why would they need it on any other electronic device designed for the same environment?

    ReplyDelete
    Replies
    1. SW,
      Good point. I've seen it pointed out that commercial FM radio is going to be affected.

      There other systems are way more important.

      We really don't want to be blanketing our airwaves with lots of Radio Noise to preserve a TELSTRA asset. If the Government still owned Telstra, it might make sense, maybe.

      That it isn't our public asset any more kills the economics.

      cheers
      steve

      Delete
    2. In my suburb Barrack Heights NSW there are no pits except on the main road so every Node would have to be above ground 10 in my street alone!

      Delete
  4. If we're comparing "bleeding edge" technologies copper's G.fast promises up to 1Gbps over 200m but this diminishes into insignificance when compared with "bleeding edge" fibre technologies - 69.1 terabits per second over a single 240 km long optical fiber (http://phys.org/news189430420.html) and even better at NEC with 101.7 terabits per second through 165 kilometres of fibre (http://www.newscientist.com/article/mg21028095.500-ultrafast-fibre-optics-set-new-speed-record.html#.UgwkP00ibaF).

    It's senseless to spend $29B to upgrade end-of-life copper when fibre even in it's infancy offers vastly better performance now and new developments using the same fibre demonstrate it's potential in the future.

    For $37B future proof fibre should be considered a bargain over the LNP's $29B throw away copper plan.

    ReplyDelete
  5. Nev,

    Well said. 1925-spec Copper really isn't a good base for a 21st Century business...

    cheers
    steve

    ReplyDelete

Note: only a member of this blog may post a comment.