back to article Why 'slow light' might just save the Internet

If we want to keep expanding the performance and the reach of the Internet, we need an inflection point: otherwise, its electricity consumption will become catastrophic. This isn’t some hippie greenie Luddite concern: even if we ignore the climate question, we need to be able to supply our telecommunications networks with …


This topic is closed for new posts.
  1. another_vulture


    You started by defining "the Internet" to ionclude its servers, and you are absolutely right. Buyt hte server component is siliocon and electronic, not photonic,. For servers, the figure of metis as recently as 2005 was ops/$. Now, the figure of merit is ops/Watt. Internet content providers (Google, Facebook, etc.) are no longer compute-constrainted, so now the cost of operatins is driven by the energy used. We can expect ops/Watt to continue to decrease even in the pure electronic domain due to increased integration at the chip level. Later, we will start to see power reduction in servers at the board level when chip-to-chip photonics replace chip-to-chip electrical signalling.

    At the data center level, we will see much more power-efficient LANs. Up to now, LAN technology was driven by bps/$. But now, we can also look at bps/Watt. and dramatic improvements are possible.

    One upshot of all of this is that the capability of a physical data center (ops per cubic meter) will continue to increase exponentially, even though our old metrics such as CPU cycles hve flattened.

    1. Anonymous Coward
      Anonymous Coward

      Re: Servers

      > we will see much more power-efficient LANs

      I hope so. I had a peek at the nexus power consumption and it said 90w/port.

      I could have got the wrong end of the stick, but that seemed outrageous.

    2. Circadian

      Re: Servers


      way to completely miss the point of the article. Your optimistic "...will continue to increase exponentially" is about to run into a brick wall called "physical limits of the current technological process". Which is the whole point of this article - scientists and engineers trying to find new technologies to replace those that are about to hit their limits.

  2. h3

    Wrong Day

    I am far too drunk to read that properly.

    1. Alan Dougherty

      Re: Wrong Day

      Yep, started reading, looked at empty wine bottle, cursed Canadian F1 drivers again, thenn I think I goingto ebd..

  3. Anonymous Coward
    Anonymous Coward

    +1 for honesty and fortitude

    see above

  4. Anonymous Coward
    Anonymous Coward

    Will any of this make my bt home hub 3 maintain a reliable wifi network

    for longer than half a day?

    1. fearnothing

      Re: Will any of this make my bt home hub 3 maintain a reliable wifi network

      That would require BT's sourcing people to start selecting hardware/software that isn't shit.

      So no.

  5. Disgruntled of TW
    Thumb Up

    ... so, does the 50V we in the UK currently shunt down every copper telephone line to power the ringer, heating the insulation with16.46 Ohms per 1000ft at 25C, help balance the power budget required to send wavelengths down a fibre? Do any of them have some numbers on what this requires in power budget?

    The discretionary power usage in a datacentre varies far too much according to the whims of shareholders rather than communications necessity. I believe there are a great many years left in our "essential" datacentre power budget before we need optical switching - which is just as well, as it could take us a fair few years to get it working commercially.

    The Ozzies do talk straight, but I'm missing a few variables in the equations they're producing.

    The optical flip-flop would be a nice toy. Yes please.

    1. Nuke

      Disgruntled of TW

      Wrote :- "... so, does the 50V we in the UK currently shunt down every copper telephone line to power the ringer, heating the insulation with16.46 Ohms per 1000ft at 25C"

      You must be using an antiquated phone if it takes its ringer power from the line and, if it does, that it takes a significant current. Is it black, weighs about 4Kg, and has like a pair of bicycle bells on the top by chance? Even so, it doesn't spend all its time ringing does it ? (Oh, perhaps it does).

      You mention 50V as if that is an efficiency minus point, but in fact the higher the voltage the more efficient the transmission is. That is why the National Grid runs at typically 500,000V. The 50V is not "shunted" down the line; it is a charge. It is current (as in Amperage) that is shunted down the line. I have quite a few close encounters with my overhead line when I prune my trees to clear it, particularly like today after the wind has broken branches, and there is no way it is at 25C.

      1. Martin 71 Silver badge

        Re: Disgruntled of TW

        1. Black phones tend to have the bells inside, the ones with them on top were wooden

        2. The 50v is to feed the DC speech circuitry, ringing voltage is 75-80v at 16-25Hz

        3. The national grid is a mix of (mostly) 400kv with some smaller segments still on 132kv and 66kv, not 500kv

        4. The 25degree C figure is the temperature at which the resistance is measured, not the temp it's alleged to be heated to by the current.

      2. John 62

        Re: Disgruntled of TW

        I finally got a pair of el-cheapo DECT phones last year, but kept the line-fed phone plugged in, too. Handy if the power goes off and you need to, say, phone 999 or something like that.

    2. Martin 71 Silver badge

      Not much I wouldn't think

      Most modern exchanges feed a set current down the line (20-30mA if I recall, can't be bothered to stick a DMM into the phone socket to find out), but the current demand of modern systems is TINY compared to their electromechanical ancestors (where the selectors usually took ~1 amp when stepping).

      The actual current consumption of most phonelines is much smaller now, as people tend to use their mobiles more than the landline. The non-use leakage of the line is tiny, less than 2mA, as around 2-3mA will trigger them to send Openreach round randomly to bill you 100+ quid for finding a fault that's not there.

      1. Anonymous Coward
        Anonymous Coward

        Re: Not much I wouldn't think

        There seem to be two constraints on taking current from a BT landline. One is indicated by the REN to reliably ring a small number (4?) of extensions/devices. The other is the number of CLI displays - which is/was limited to something like 2 or 3 connected units.

      2. ja

        Re: Not much I wouldn't think

        We are moving into town and the land-line, TV and IP connection will come in on fibre. The Canadian telco's are really pushing fibre to the home these days. I believe the Americans are doing that as well. What are you up to in Blighty and Europe? Is this another UXO opportunity or perhaps make-work for archeologists?

        We can't be that short of green electricity as the big data centre companies aren't falling all over themselves to locate near sources of vast land-locked supplies of hydro-power. I haven't seen anyone offering to build data centres in Labrador yet and there is at least 2,000 MW of undeveloped capacity at Gull Island on the Churchill River. It will be far cheaper to operate in Labrador as the major expense is transmission rather than generation and the ambient temperature is several degrees cooler than New England or Europe. If a customer were to get involved early enough in the design, it should be possible to dedicate multiple generators at multiple dam sites to supply power. That should reduce or eliminate the cost of dedicated UPS. Labrador is remote yes, but it is only a few hundred kilometers from major trans-atlantic cables and it lies close to great circle tracks from the US mid-west to Europe.

      3. Alan Brown Silver badge

        Re: Not much I wouldn't think

        Eelectromechanical systems had power consumption directly proportional to the number of circuits actually in use. Electronic ones draw near constant power.

        Stepping current only pulses for 1/20 of a second. Electronic switches drew considerably more aggregate power than the mechanical switches when first introduced.

        30-50ma current loop design dates back to carbon microphones. I'ts amazing how long it's lasted.

  6. Charlie Clark Silver badge


    It's arguable that the internet (and it certainly shouldn't be capitalised if it is being used as an "omnibus" term) is displacing other even more energy hungry activities such as physical mail. Be that as it may, growth is now increasingly wireless so it's difficult to see advances in optics making much of a difference there - yes, I know backhaul and data centres are still cable-based, but the article's premise is "growth".

    The point about roof-panel solar energy depressing peak electricity prices is spurious: the spot market is irrelevant for data centres. You might want to compare growth in absolute terms of cheap renewable power generation with consumption by data centres and devices, though the increasing fungibility of gas is currently driving prices.

  7. Kevin Pollock

    A few corrections

    >>"One of the best ways to cut down networks’ power consumption is to get rid of the power-hungry electronics that does most of the heavy lifting."

    It's not quite as simple as that. Service speeds tend to be a lot lower than backbone data rates (especially as DWDM moves towards coherent super-channels). So you *can* build all-optical networks, and they may appear to have a lower CapEx and power consumption, but they will be inefficiently filled unless you can do electronic grooming of services in the core network. There are several mathematical studies of this if anyone's interested.

    >>"Optical communications is based on the 1530 nm wavelength band – the entire ITU frequency grid for DWDM systems (G.694.1) fits between 1530 nm and 1625 nm – a tiny amount of the near infrared (the vertical red line on the graph above)."

    No. Let me see if I can rewrite this to keep it both accurate and concise.

    Optical communications is based on a range of near infrared wavebands, including the C-Band (1530-1565nm), and the L-Band (1565-1625nm) - both of which are used for DWDM long haul optical transmission. Other wavebands (O, E, S Bands) are used, along with the C and L bands, in Coarse WDM transmission.

    >>We don’t use visible wavelengths, because the ubiquitous Erbium-doped optical amplifier, cheap and everywhere, works at the 1530 nm band.


    We don't use visible light because the attenuation of silica-based fibers at those wavelengths is way too high - whether you have workable amplifiers or not. Multimode, lower data rate LAN systems make use of red LEDs, but only for very short reach applications.

    You are correct that EDFAs work best in the C-Band - in fact they "define" the C-Band. EDFAs can also be made to operate in the L-Band (but they are different EDFAs). Because of the energy levels in the Erbium atom, EDFAs do not work in the O, E and S Bands.

    Semiconductor Optical Amplifiers (SOAs) work in all of these bands, but there is a long-held perception that SOAs are not suitable for WDM operation. EDFAs are defintely better - but the fact is that optical attenuation outside of the C and L Bands are generally too high for high data rate, long haul transmission.

  8. John Smith 19 Gold badge

    In RF 5% of the maximum frequency is *not* broadband

    An Octave (50%) *is*

    Optical has a *long* way to go in that area.

    The 3rd harmonic stuff is interesting as up-shifting the frequency normally implies very *low* efficiency. IIRC in the watt power level rather than the mW.

    Remember folks another article in this series mentions internet traffic is growing by 40% a year so somethings got to be done.

    Of course 39.99% is a)Spam and malware b)Pron.

This topic is closed for new posts.

Biting the hand that feeds IT © 1998–2021