IEEE Seeks For Ethernet To 'Go Green'

alphadogg submitted a piece at the NetworkWorld site about the IEEE's efforts to introduce energy efficiency to Ethernet use. The group's Energy Efficient Ethernet group is looking into methods by which standards can be tweaked to encourage power savings. Current plans include ways to make computers 'choosier' about what level of bandwidth they're using. Idle systems would only run at 10Mbps, while email might draw 100Mbs, and scale up to 1000Mbps for large downloads and streaming video. The group is planning to discuss changes to the Ethernet link and higher layers. No restrictions are planned for device manufacturers, although the article suggests some companies might try to use energy efficiency as a competitive advantage. The EEE group estimates some $450 million a year could be saved via the use of energy efficient Ethernet technology.

Comments

Saving energy now

[jbeaupre]

Seems they are saving energy by throttling bandwidth for the article. Any manage to read it?

Acronym confusion?

[MECC]

Once Apple adds the ability to negotiate EEE in Macs, they'll call it iEEE.

Re:Acronym confusion?

[mrfantasy]

It's the IEEE Energy Efficient Ethernet Initiative.

Or, the IEEEEEEI.

Commonly pronounced "eye-six-N-eye".

Re:Saving energy now

[jbeaupre]

Nevermind ... I just had to try connecting 3 or 4 times. Interesting idea. Let's see ... throw out millions of PC's with integrated ethernet, replace them with new machines. Oh, guess they mean in a decade or so through normal replacement.

I have an idea

[Salsaman]

Use more zeros and fewer ones.

Re:I have an idea

[ToxikFetus]

On an open collector data bus, '1's would actually uses less power since that is the high impedance state. The '0's pull down the current.

Re:I have an idea

[skiingyac]

If you really wanted to, you could include a sync preamble like is done in many wireless physical layer protocols (might negate any efficiency gains though), or use an encoding like is used in CDs to ensure you don't end up with too long a string of all 1's or all 0's that the clock drift/differences cause ambiguity (less efficient than no encoding, but possibly better than manchester).

Power over Ethernet Could Help

[Kadin2048]

One of the easiest ways that the Ethernet people could encourage energy efficiency would be by promoting greater use of Power Over Ethernet. By moving networked devices away from each having an individual wall wart, which are typically inefficient (as well as inconvenient), PoE lets you concentrate the AC to DC conversion in one place, for greater efficiency. As long as you don't have terribly long cable runs, I think there would be a significant net savings overall.

The number of networked devices people are going to have in their homes is only going to grow. I think a big segment could be in "Micro NAS" devices, basically single HD boxes that plug in to a home network and add storage that's accessible from any computer in the home. They're smaller and cheaper than RAIDed NAS solutions, but more convenient for people who have multiple computers than a FireWire or USB2.0 hard drive. And then you have routers, WiFi APs, network cameras, set-top-boxes for playing back video and audio, etc. All of those light-draw devices could be powered over the network connection instead of each having a wall wart.

Re:Power over Ethernet Could Help

[GigsVT]

Running power over tiny 24 gauge wires is very inefficient too. Try again.

Re:Power over Ethernet Could Help

[p0tat03]

An idea I've always thought about is converting to DC supplies indoors. AC has an advantage in terms of long-distance transmission, but in this day and age a HUGE part of our electric use is in devices that require DC power. Hell, many of the things that run AC (like lights) can in fact run DC with nary a problem. It's always boggled my mind why we have a bajillion power bricks sitting around, each venting heat like mad converting AC/DC, when in fact we could have a much more efficient "main" transformer installed in the house that does it on a larger scale and feeds our devices directly.

I imagine this would be even more useful for heavy power using environments like server farms - imagine if you can do with the huge boxy PSUs in every single box and just have a unified DC power source that can FAR more efficient than what's in the average beige boxen.

Re:Power over Ethernet Could Help

[NekoXP]

Just because it doesn't generate heat doesn't mean it isn't losing power. The energy wasted in relation to the power on the cable is probably quite high (DC doesn't travel well, that's why wall power is AC remember) compared to just wiring up wall sockets and using warts or switching PSUs.

You're just transferring the wall-wart to another room though, and making the loss over the cable add to the power inefficiency. Imagine the extra airconditioning provision the room with the new site-wide AC-DC converter will need :D

PoE is a clever way to power devices that are in hard to power places (where you can wire a network using a thin cable but far away from a power socket) and keeps devices cheap (no need to do anything but DC-DC conversion from PoE to components) but it's not any better energy-efficiency-wise.

Can't this IEEE stuff they're talking about simply be built into drivers? I know my laptop ethernet (Intel) has the ability to scale down the ethernet speed when the battery is in use, or during standby and so on. Would it cause too much trouble to have the driver anticipate and schedule a renegotiation on a power source change or based on activity? Why would ethernet vendors need to be involved if it was simply a driver 'problem' - apart from having to write drivers that do it for their hardware (which most of them DO already).

Can't we have a sysctl or a sysfs tweak in Linux/BSD/whatever to demonstrate it and see if it even helps? Does networking hardware at the other end (for instance a 32-port Cisco switch) actually use less power if half it's ports are at 10mbit rather than 100mbit?

Can't we do this with wireless? 802.11b etc. already has power calibration built in but could it pull it back when the bandwidth requirement isn't so high, saving battery life and not polluting the airwaves with high powered chatter? My card uses the same transmit power whatever the state of the laptop is..

Re:Power over Ethernet Could Help

[Kadin2048]

An idea I've always thought about is converting to DC supplies indoors. AC has an advantage in terms of long-distance transmission, but in this day and age a HUGE part of our electric use is in devices that require DC power. Hell, many of the things that run AC (like lights) can in fact run DC with nary a problem. It's always boggled my mind why we have a bajillion power bricks sitting around, each venting heat like mad converting AC/DC, when in fact we could have a much more efficient "main" transformer installed in the house that does it on a larger scale and feeds our devices directly.

I imagine this would be even more useful for heavy power using environments like server farms - imagine if you can do with the huge boxy PSUs in every single box and just have a unified DC power source that can FAR more efficient than what's in the average beige boxen.

It is a good idea; in fact it's such a good idea that people have been thinking about ways to try and implement it in datacenters for a while. Unfortunately one of the bigger problems is that most motherboards don't run off of a single voltage; they have +5, -5, +3.3, +12, and so on. There has been a push by some big server-farm operators, Google in particular, to encourage board makers to produce mobos that only require a single +12V supply, because then you could do exactly what you say: have a big AC to DC converter somewhere (probably running from a medium-voltage AC main) and then distribute the 12VDC around to the racks.

It was a Slashdot article back in September:
http://hardware.slashdot.org/article.pl?sid=06/09/ 26/2039213

Re:Power over Ethernet Could Help

[Shakrai]

(DC doesn't travel well, that's why wall power is AC remember)

This is a very common misconception. Low voltages don't travel well because you need more current (i.e: amps) to carry the same amount of power and this requires bigger wires. The main reason your wall power is AC is because it's easier and cheaper to build transformers for AC that convert high voltages (for distribution) into low voltages (for usage).

DC is actually used in electrical distribution. It's known as HVDC and it's actually more efficient then AC because it doesn't have to contend with capacitance issues.

Re:Power over Ethernet Could Help

[dattaway]

just use zener diodes to get the voltage down to whatever it is that the device requires?

Using shunt regulation? Bleeding off what you don't use in the form of heat? That's worse than linear voltage regulators!

Re:Power over Ethernet Could Help

[Ungrounded Lightning]

It is a good idea; in fact it's such a good idea that people have been thinking about ways to try and implement it in datacenters for a while.

Actually the networking industry DOES do it that way. SPower supply to many routers (such as ALL the ones some major companies make) and other networking gear is redundant 48V DC - a standard for networking equipment dating from the days of relays. (Line powered units have extra line powered supplies to make the 48 DC.)

Not only that, but often the boxes don't have a per-box 48-to-whatever supply. Instead each blade requiring other voltages has its own switching regulators.

(This isn't just for efficiency - it's also for redundancy. A box power supply is a single point of failure for the box. Give each card its own supply running directly from the redundant power busses and if one fails all the other cards in the box keep working - meaning only the lines to that card are in trouble, not everything hooked to the box. You have to pull a card with a failing component to replace it anyhow - so if you want to cover the lines to it in case of card failure you need other redundancy anyhow. So single points of failure on a card are OK.)

Power requirements on modern ASICs, networking processors, and RAMs are getting higher, operating voltages lower (for better speed-power products) forcing higher currents, and switching DC-DC converter/regulators are getting more efficient. These days it actually makes sense to add an additional regulator near a major load so the power can cross a few inches of the PC board at a higher voltage and lower current, to avoid heating and voltage droop in the layer of copper that carries it. You're starting to see that in PC motherboards, too.

Re:Power over Ethernet Could Help

[D4rk Fx]

Can't we do this with wireless? 802.11b etc

Somehow I think Power Over 802.11b would prove quite difficult to implement....

Re:Power over Ethernet Could Help

[PitaBred]

The 50V limit was actually put into law by telecom lobbyists. Under 50V, you don't need an electricians license, and they wanted to be able to do telecom work without electricians. 50V, and even 25V can be quite dangerous if you don't pay attention.

Green

[truthsearch]

IEEE Seeks For Ethernet To 'Go Green'

That's good because I'm really tired of the white and blue.

Re:Saving energy now

[skoaldipper]

I did. The problem (FTA):

"One challenge is finding a way to make a PC or laptop network interface card (NIC) change gears more quickly -- "a couple orders of magnitude faster than auto-negotiation, to make the switch as seamless as possible," Bennett says. "Auto-negotiation runs at about 1.4 seconds and we're talking about -- just to start the discussion -- a millisecond of switching time."

So, why not just set NIC(s) to negotiate at the lowest speed first? Then throttle up gradually based on end to end transmission intervals. They talked about using buffers and NIC electrical consumption to handle the negotiation. I say, just start at 10mbps and negotiate up to Gig speed gradually, and make the firmware drivers allow one to turn that feature off/on and back to our current default. My simpleton mind must be overlooking something.

Well Duh!!

[eclectro]

Another suggestion - Stop all the spamming. There must be a coal-powered powerplant's worth of electricity right there.

What about the power supplies...

[russ1337]

Sounds like someone is really starting at the wrong end. IMHO.

I'd estimate that power supply inefficiency chews up more than this proposal will ever save. If you spent your time making the power supplies of PC's, Switches, routers more efficient you'd probably have a greater impact. How about better efficiency in the FET's, transistors and amplifier circuitry? Last time I checked, my Ethernet looms didn't get that hot. (isn't it all about "(i^2).R"?. Heck turning off the light in the switch room probably does more to save power. Plus all the heat im my server room is from the servers, not the Ethernet. If your that worried, switch to fiber.

I thought the transfer of data at the physical layer was through the transfer of 'holes' anyway.

Re:What about the power supplies...

[zaf]

Exactly. All the hundreds of devices independently converting AC voltage to DC all day long is far more power waste than what's inside the CAT5. Speaking of, whatever happened to the push for DC datacenters? As far as I can tell, there's still no widely-used DC standard as an option for most of the devices in a small-medium sized environment

Re:What about the power supplies...

[jaredmauch]

When you're talking about larger switches and routers and not the cheap linksys/dlink crap most people call a "router, there was actually a good presentation at NANOG last year. You can watch it(real video) from the link (and view slides). Most of the efficency in these larger devices has already been done. (obviously excluding that whole google + pc power supply discussion). Check it out if you are truly interested in this space.

Question?

[SnarfQuest]

Does 100 (or 1000) really take that much power to download one "file", or is it the same amount of power used, just in a shorter time period?

Or is it power used while idle? Does a 1000 device comsume more power idling in that mode than a 10 device would?

Re:Question?

[Ant P.]

I think what they're proposing is clock frequency control for Ethernet chips, like CPUs have now. I read somewhere that the power consumption increases n^3 with the clock speed, dunno where that figure comes from though.

Re:Question?

[greed]

It's the square of the clock speed; it comes from some math in second- or third-year Electrical Engineering.

It has an awful lot to do with line capacitance and inductance; you've basically got to "fill up" the line before you can see the signal change at the other end. (Be it at chip-level or network-cable-level.)

Which is why narrower fab processes and low-voltage differential signaling is so important in high-speed circuits; all those watts are heat that has to be dissipated. Narrower CMOS gates take fewer electrons to charge up. And by also reducing the voltage needed to see the signal change, you can reduce the impact of that clock speed increase.

But that also means the old, slower speeds with modern signaling could be run on nearly no power. Which we do; that's how the iPod and cellphones get smaller and runs longer each year. (Dropping analog support on a 'phone helps a lot, too.)

Measurable?

[zaf]

Does that much current actually go over ethernet transmissions? It seems to be that more power could be saved by more efficient power supplies in the switches than by wasting a lot of time and research in figuring a way to throttle link speeds. Does anybody have a value for the amount of electricity used for an hour's worth of data at 10 megabits as opposed to 1 gigabit?

It just surprises me that +/-5 volts over copper really makes all that much difference compared to all the other waste in the datacenter.

Also, what's the difference in energy usage for copper vs fiber links??

Re:Measurable?

[Matt_Bennett]

Actually, it is pretty surprising how much current Gigabit takes- The output drives usually work in a current mode, and they draw 40mA per pair- since gigabit uses 4 pairs, that's 160mA on each end of a gigabit link. *But* the big difference is in what happens when the link is idling- 10mbit only puts through link test pulses, but 100Mbit and Gigabit both keep up idle patterns that are basically encoded strings of no information- this keeps both ends of the link ready to accept data- 10Mbit has to transmit a synchronization series of pulses to make sure both ends are clocking at the same rate. For 100 and gig, at least to the output drivers, they draw the same amount idling or transmitting at line-speed.

They don't like orange?

[bkoehler]

I see no reason to switch from 568B to 568A.

By any chance...

[ivan256]

...is this group led by ethernet equipment vendors? Perhaps vendors who are unhappy with the recent decline in equipment upgrades since people aren't upgrading from gigabit or even to gigabit from 100mbit in a way that helps their stock price sufficiently?

It seems to me that, considering the number of ports active out there, they're talking about a tiny amount of savings per port for a total investment that could have a much larger effect if spent elsewhere.

Hell, I bet more power is wasted by the power supplies, overly conservative fan controls, uncleaned air filters, shorted out UPS batteries that should have been replaced decades ago, overpowered CPUs, and crappily written firmware of the currently deployed switches than is consumed by transmission losses.

Re:Saving energy now

[Anonymous Coward]

For broadcom Ethernet PHY chips, they use about 1W/port when configured as 1000BaseT (GigE). GigE require some heavy duty DSP filtering as well as driving 4 pairs of bidirectional transceiver. They would burn less power when they are running at 100BaseT which only to drive 1 pair of receive and transmit. Not sure if there are significant saving going down to 10BaseT as the number of transmit pairs and the DSP's are dominant.

While this might not seem a whole lot of power, when you are looking at Enterprise size (say a few hundreds to thousands ports) setup, there can potentially be savings at the few hundred watts to thousand watts range.

I recall the first time I noticed ethernet power

[erroneus]

At an office I once worked, there were a lot of spare switches laying about after upgrading to 1000BaseT. They were considered "spare" or whatever, but there was a great many... so I sorta brought one home and mounted it into my rack and used it for a couple of months. The next two electric bills made me rethink how nice it looked to have a 24 port switch in my rack instead of that cheapy 8 port sitting on a shelf. It consumed a NOTABLE amount of power. Now, there were other things involved I'm sure... things like the changes of the seasons, global warming and all that. But when I brought the switch back to the office and went back to my cheapy 8 port again, I saw a change in my power bill.

If I ever decide to spend money on a nice looking switch, I'll be sure to reference the power draw of the units I review.

Patch cables are now easier than ever!

[PainBreak]

So a straight-through is: Green white / Green / Green white / Green / Green white / Green / Green white / Green Sweet. Crossovers then would be: Green white / Green / Green white / Green white / Green / Green / Green / Green white So much easier to remember! Thanks, IEEE!

Two methods

[HomelessInLaJolla]

> idle or underutilized Ethernet connections more energy efficient

There are several ways to increase measured efficiency. Two of them include:

1) Load the network with verbose transmission protocols, junk, or spam such that more network cards have higher sustained traffic (quantity means more than quality from the usage point of view).

2) Increase the number of hardware exploits such that underused network adapters can be continually used by those who know of the hardware exploits (make the network adapters available to those who have convinced themselves that they need more bandwidth than they're willing to pay for)

This is _not_ a troll. :)

It's not the power at the physical layer -

[wsanders]

- it's the power required to process the packets. More or less, a GigE card should need 10X (divided by some fudge factor that probably makes the real ratio closer to 2 or 3X) the compute power of a 100Mbit card. Processing GigE at full throttle actually takes quite a bit of CPU - we don't notice it much because most GigE interfaces have a TCP Offload Engine that avoids bogging down the CPU and bus.

So your TOE could easily have a variable speed CPU that basically goes to sleep when it can negotiate the physical interface down to 10Mbit, or whatever. SOunds pretty straightforward.

Hmmm....

[NerveGas]

    Don't get me wrong. I'm all for being green. But it would seem that instead of putting all of that effort, design time, and eventual costs in equipment in order to save a very small number of watts on the ethernet chips at each end of the link, a slightly larger effort directed into power supply losses, CPU power usage, or GPU power usage would yield 10x the benefits.

    Realistically, I know that they can't just walk over to Intel, AMD, and NVidia, and say "Alright, guys, we're here to tell you how to use less power." They're just doing what they can, and they deserve applause for it.

Re:Saving energy now

[phayes]

Given the number of times that autonegotiation has given me headaches because supposedly compliant devices couldn't agree on how to setup a connection, I wouldn't want to set this up on any of my networks. I just can't see myself explaining to the CIO that the reason that the ERP is slow to the point of being unusable is because the core switches renegotiated their bandwidth down to 10Mbit/sec overnight when they were unused and were unable to ramp it back up again correctly. There is a reason that autonegotiation is often disabled & it's called experience...