Forgot your password?
typodupeerror
Cloud Power Hardware

AC and DC Battle For Data Center Efficiency Crown 168

Posted by samzenpus
from the who-made-who dept.
jfruh writes "AC beat DC in the War of the Currents that raged in the late 19th century, which means that most modern data centers today run on AC power. But as cloud computing demands and rising energy prices force providers to squeeze every ounce of efficiency out of their data centers, DC is getting another look."
This discussion has been archived. No new comments can be posted.

AC and DC Battle For Data Center Efficiency Crown

Comments Filter:
  • Makes sense. (Score:5, Interesting)

    by Anonymous Coward on Wednesday March 14, 2012 @04:30PM (#39357313)

    AC is better than DC for transporting electricity because you can convert between voltages with just a transformer. But in a data centre, when all the equipment will be powered by the same voltage, it makes sense to use one good efficient power supply for multiple computers, so that all the components don't have to be duplicated for each computer.

    • by Skapare (16644)

      Yes, let's use big power supply for all computers, so they all share the same exact point of failure AND have a MASSIVE fault current when someone accidentally drops a piece of uninsulated wire across a bus bar, so we have have a couple racks of equipment meltdown and a techie vaporized to ash.

      • by vlm (69642)

        Thats why, like, I donno, 80 years ago, the telco business got in the habit of A and B power bus distribution. I worked at a place with a C bus which was pretty much a load balancing hack and confused the hell out of the CO techs and electricians... They actually shorted out the C bus one time because they didn't understand the concept of having three busses instead of the "standard" two.

    • Re:Makes sense. (Score:5, Informative)

      by betterunixthanunix (980855) on Wednesday March 14, 2012 @04:42PM (#39357477)

      AC is better than DC for transporting electricity because you can convert between voltages with just a transformer.

      Which was a winning argument in the 19th century, but not anymore. The use of AC entails significant power loses, especially for cables that are immersed in salt water, which is why DC is used in such situations:

      https://en.wikipedia.org/wiki/High_voltage_direct_current [wikipedia.org]

    • Re:Makes sense. (Score:5, Informative)

      by tlhIngan (30335) <(ten.frow) (ta) (todhsals)> on Wednesday March 14, 2012 @04:43PM (#39357483)

      AC is better than DC for transporting electricity because you can convert between voltages with just a transformer. But in a data centre, when all the equipment will be powered by the same voltage, it makes sense to use one good efficient power supply for multiple computers, so that all the components don't have to be duplicated for each computer.

      It depends.

      AC wins out because of ease of conversion, becaues the higher the voltage, the lower the current, and lower the current, the lower the IIR losses in the wire. DC didn't win because at the time, efficient (and cheap) voltage converters didn't exist. These days, a switching DC-DC supply can easily exceed 90% efficiency, and you can get solid-state converters that can handle transmission line powers easily. Hence the launching of HVDC transmission lines which don't have resonant losses and no phasing issues

      In a datacenter, you'd probably take the incoming power and turn it into an intermediate voltage like 48VDC per rack or something - something that minimizes IIR losses (you want high voltages) and DC-DC converter losses (ideally you want output voltage and no converter).

      It will have to be per-rack at the minimum purely because of the losses - if we did 12V lines and a few servers take 1200W total, we're talking 100A in current If we bump it to 48V, we're dealing with 25A (maybe 30A after inefficiencies), and IIR losses at 25A are lower than at 100A (it increases with the square of the current).

      Also, the 100A cables are big and chunky (which you need because they reduce the "R" part of IIR losses).

      • Re:Makes sense. (Score:5, Informative)

        by effigiate (1057610) on Wednesday March 14, 2012 @04:55PM (#39357633)
        One of the challenges of HVDC, especially in the transmission/distribution world, is that normal switching happens on the line and not at the breaker. If you can switch futher down the line, you can leave all the people closer to the breaker with power. The issue is that this switching happens while current is flowing which requires that the device interrupts real current. In the AC system this is relatively easy because the arc created by opening a high voltage circuit under load goes out at every current zero. There is no current zero on DC, so you force the interrupting device to break current. An similar situation can be seen if you look at relay contacts. They may be rated at 20A @120VAC but only 0.5A at 12VDC.
      • by Idbar (1034346)
        Well, my way of thinking is that, what doesn't make sense at all, is that normally in data centers, you convert AC to DC, then to AC again, then to DC again.

        Yes... many data centers these days have UPS. So my guess is that should be more efficient to get the DC right from the batteries somehow to avoid the losses of having those two extra conversions. So agreeing with your comment, I assume it would be efficient to have small UPS systems (rectifiers and batteries) per rack (or per small group of racks), t
    • The problem is it wasn't back when the grid was being made. There was no good, easy, efficient way to convert DC voltage. Now, not so hard.

    • by Chrisq (894406)

      AC is better than DC for transporting electricity because you can convert between voltages with just a transformer. But in a data centre, when all the equipment will be powered by the same voltage, it makes sense to use one good efficient power supply for multiple computers, so that all the components don't have to be duplicated for each computer.

      Unless you want to transmit with lower loss and send more current down the same cable. That's why high-voltage direct current [wikipedia.org] is used for most undersea cables

    • by vlm (69642)

      AC is better than DC for transporting electricity because you can convert between voltages with just a transformer.

      Not anymore. The greenies / cost cutting / etc means no more xfrmrs anymore. Bye bye to that technology. Whens the last time you bought a wall wart charged device with a transformer inside it (you'd know, it'll be cubical and heavy)? You have to be pretty old by /. terms to have bought a main desktop computer without a switcher, like early 1980s era pre-PC "home computers"... Ahh the old Altair with its smoking hot 7805 regulators...

      Since you're gonna have a switching power supply anyway... why not ski

      • DC/DC converters are better than transformers in almost every way. They are lighter, smaller, and cheaper. Also, altought theoreticaly you could create a transformer that loses less power than a DC/DC conversor, in practice nobody did that, thus they also waste less power.

        They would be even better (on all variables above) if they didn't need to deal with a low frequency AC supply. Either a high frequency AC or a DC one would do.

      • Re:Makes sense. (Score:4, Informative)

        by Nethead (1563) <joe@nethead.com> on Wednesday March 14, 2012 @06:44PM (#39358635) Homepage Journal

        The last two data centers (Clearwire) I built out were DC. The only AC in the cage was for a video monitor and for the tech's wifi router. Very standard stuff, the telcos have always done it that way. Any bit of Cisco/Juniper/whatever kits can be ordered with DC power supplies. I see DC plants as more the standard now. And yes, the are still built using waxed string.

        Even Power over Ethernet has it roots in telco -48VDC power. All the WAPs and fiber converters at a Lowe's are powered by a Valare DC power supply ( http://www.power-solutions.com/dc-power-systems/eltek-valere.php [power-solutions.com] ).

        One nice thing about DC plants are the power cables are cut to length so you don't have all that extra line cord to bundle and hide.

      • by russotto (537200)

        Since you're gonna have a switching power supply anyway... why not skip the pesky rectifier diodes and feed in raw couple hundred volts DC? Quite a few PC power supplies work just fine off raw DC on the supposed "AC input"... good luck figuring out which work and which dont without some smoke events.

        You can run them on DC, but it's a waste; you still get the diode drop and half your diodes are going to get hot as hell; normally you have a full-wave rectifier where each diode operates at 50% duty cycle. Wit

    • whichever one gets Dirty Deeds Done Dirt Cheap

    • by geekoid (135745)

      but that's the proble. we aren't talking about 1 rack, we are talking about 10' of thousands of machines. How is the efficiency to the end of all the computers? what about heat? risk?

      I honestly don't know the answer, and I look forward to datacenters data.

    • Easier, not better. To carry the same power with the same losses, you need 1.4x more cross section in your cables. That's if you have perfect power factor. If you have poor power factor you need much thicker cables. Transformers are also not the most efficent at low frequencies like 50Hz. They require massive amounts of steel and copper too. Modern switch mode power supplies (like those used in computers, since transformers could not be made small enough the provide the required power) need to convert the A
    • While it is true that AC can be voltage converted with nothing more than a transformer, it isn't really relevant: the old school AC transformer units have miserable efficiency and are both heavy and bulky. Basically all modern equipment is going to be using a switchmode power supply that is a great deal closer(in terms of complexity, cost, efficiency, and theory of operation) to a DC-DC converter.

      Either way, you get to play the "let's balance transmission losses vs. redundancy vs. efficiency vs. componen
    • All DC in a large data centre and you'll need whopping thick bits of copper and equipment that would look at home next to a 500MW generator.
      A rack basis is one thing, and can work well. A big roomfull is another.
  • by Anonymous Coward

    which means that most modern data centers today run on AC power

    Only if you ignore all telecom equipment which have run on -48VDC for decades. True, they're not really 'data centers' but it's not like they don't use massive amounts of electricity.

    • by Skapare (16644)

      48VDC also means a rather large amount of current. A data center in many cases these days is much, much larger than a telco switching center was (aside from maybe a few trunk points for large cities). They did, in many cases, divide up the electrical systems to avoid high fault currents. But it was well know the high battery currents involved could be a disaster if there was a short, even on a branch tap into equipment.

      The benefit of DC distribution was NOT efficiency. They did use resistors and in some

  • most telephone exchange and related transmission hubs use DC 12, 24 and 48VDC are standard. This isn't anything new, and data centers have always been space and power inefficient, it's the nature of the beast, and method of construction.

    • Along with the specialized telecom equipment, a few standard server vendors, including Intel when I was there, have models designed with 48VDC power, along with NEBS-compliant features like - not catching fire.
  • by mlts (1038732) * on Wednesday March 14, 2012 @04:42PM (#39357467)

    There was an article about using 380 volts a couple weeks ago on /. in the data center.

    Having DC brings some benefits, mainly just needing to step down voltage and not have to rectify it smoothly with capacitors to even out the output current.

    However, there are some downsides:

    1: AC power supplies in devices tend to be more tolerant of power fluctuations. An all DC shop might completely be halted by a power surge/spike that wouldn't bother a data center on AC.

    2: DC sparks a lot when connecting/disconnecting. AC has plenty of zero-crossings a second (120 or so), so it won't make the fireworks show when plugging/unplugging. This makes switches rated for DC a lot more expensive than AC.

    3: There is no such thing as a NEMA 380VDC connector. So, either items would have to be wired up to a bus bar similar to how 48VDC telco stuff gets, or it will end up like 12VDC with at least 5+ connectors (direct wires, cig lighter, airplane, marine connector, male/female combined connector, motorcycle accessory connector, banana plugs.)

    4: Safety. 12 VDC shocks are annoying; a shock from 380VDC will be fatal, especially because of DC's tendency to get muscles to "lock". (This is why stun fences uses AC, while kill electric fences use DC so they can keep the target locked on the wires long enough to get the amps across the heart.)

    5: Issues with wire length. AC, it isn't hard to use a transformer to deal with voltage drop. DC, that will be a lot harder.

    All and all, 380VDC seems like a solution in search for a problem. We really don't need another standard. Heck, just pointing out 120VAC in the US means I have to doublecheck if I'm dealing with 15 amps, 20 amps, 30 amps, or 50 amps, and the locking versions of each, which means six plug types and minimum wire gauges.

    • by Microlith (54737)

      1: AC power supplies in devices tend to be more tolerant of power fluctuations. An all DC shop might completely be halted by a power surge/spike that wouldn't bother a data center on AC.

      All this does is require that the conditioning for power be done well before it reaches the machines. There will be an AC->DC power supply regardless, it'll just be much, much larger and could probably supply even more resilience than a bunch of smaller power supplies.

      2: DC sparks a lot when connecting/disconnecting. AC h

      • So you'll handle it much like most hotplug PC hardware is these days, with latches and mechanical disconnects that ensure + and - are disconnected simultaneously.

        You don't want to disconnect both simultaneously. The idea is to disconnect + first and leave ground connected. The voltage across the whole component falls to ground level instead of potentially floating up to + briefly.

        This is a different problem from what the GP was talking about: When you hot-unplug a device drawing lots of DC, it starts to draw an arc. The arc will continue to draw until it gets too long to be stable, forms a big rainbow, and then extinguishes itself. The distance depends on the v

    • by vlm (69642)

      AC power supplies in devices tend to be more tolerant of power fluctuations. An all DC shop might completely be halted by a power surge/spike that wouldn't bother a data center on AC.

      Essentially you're just removing the rectifier from the power supply, putting it outside, and feeding the same old switching supply indoors. Not so. You could design a system that intentionally was more sensitive, but no one would intentionally do that.

      or it will end up like 12VDC with at least 5+ connectors

      The world seems to be converging on the Anderson Power Pole connector (which I believe is a (TM)). Cheap, high current, tough, reasonable simple to assemble...

      All and all, 380VDC seems like a solution in search for a problem

      See the above. Basically you're doing a lot of foolishness to remotely mount the rectifier diode

    • 4: Safety. 12 VDC shocks are annoying; a shock from 380VDC will be fatal, especially because of DC's tendency to get muscles to "lock". (This is why stun fences uses AC, while kill electric fences use DC so they can keep the target locked on the wires long enough to get the amps across the heart.)

      While 380VDC is really bad news, the myth that DC is more dangerous than AC is just that. A myth. In fact, AC will induce tetanus more readily than DC and cause fibrilation at much lower currents. (Given typical frequencies. High frequencies will not due to skin effect.) i.e.:

      The high voltage direct current (DC) electrocution tends to cause a single muscle contraction, throwing its victim from the source. These patients tend to have more blunt trauma. Direct current electrocution can also cause cardiac dysrrhythmias, depending on the phase of the cardiac cycle affected. This action is similar to the affect of a cardiac defibrillator.

      Low voltage alternating current (AC) electrocution is three times more dangerous than DC current at the same voltage. The lowest frequency for electrical current in the United States is 60 Hertz (Hz) because this is the lowest frequency at which an incandescent light functions. With AC electrocution, continuous muscle contractions (tetany) may occur, since the muscle fibers are stimulated at between 40 to 110 times per second. With tetany, the victim tends to hold on to the source of current output, thereby increasing the duration of contact and worsening the injury.[2]

      (http://www.medscape.com/viewarticle/410681_3)

      I've had the original Berkeley student experiments where they studied tetanus and AC vs DC, but I've lost the link. In either case, the results were much as they are reported above, i.e. it takes more than twice the DC current to "lock

  • by thsths (31372) on Wednesday March 14, 2012 @04:46PM (#39357535)

    AC, DC, it does not make a difference any more. Yes, you have to rectify AC before it powers a computer, but the rectification costs less than 1% of the energy. Power factor compensation can be more costly, but it could be avoided by going to a 3 phase rectifier. There are also serious distribution advantages in 3 phase electricity, but it is not used because of the extra complexity, despite being cheap.

    DC distribution is expensive, and 1% gain is just not enough to pay for it. Once we have intelligent grids, the situation may be different, but for now there is just no business case.

  • What if you want to electrocute an elephant?

    • Re: (Score:3, Funny)

      by KBehemoth (2519358)
      The appropriate demo of the dangers of AC data center power will be to show an elephant losing his entire database due to a power failure. Ominous voiceover: "Unlike an elephant... AC-driven data centers always forget!"
  • by ZorinLynx (31751) on Wednesday March 14, 2012 @04:47PM (#39357549) Homepage

    Standard -48VDC current distribution requires four times the current as 208V AC distribution for the same amount of power. Have you seen DC cabling at data centers that use it? If we're going to start using DC in data centers we need to come up with a higher voltage standard, otherwise we're going to spend all the savings on more copper (which is expensive!) to carry those extra amps.

  • you forgot the lightning bolt

  • If it was good enough for Edison, it's good enough for me!

  • Slashad (Score:4, Insightful)

    by Shadowhawk (30195) on Wednesday March 14, 2012 @04:55PM (#39357641)
    Articles main source says modern AC and theoretical DC are about the same. By the way, he has a product to sell...
    • by Skapare (16644)

      And you don't literally need, or need all of, his product, to make a very efficient AC-based data center.

      I am concerned about his brief mention of cooling that seemed to be based on using a single system. There, I would want multiple redundancy at N(4)+2. The more discrete units you have, the more STABLE you can hold the temperature. The more stable the temperature, the higher temperature you can run it at. UNSTABLE temperatures cause damage to equipment as much as too high a temperature.

  • by Pontiac (135778) on Wednesday March 14, 2012 @04:58PM (#39357661) Homepage

    In 2005 we started looking at blade chassis and tested a rack of HP BL series blades.

    That system came with a 48v DC power enclosure with 6 hot swap power supplies. It sat in the bottom of the rack and had a buss bar system to feed every chassis in the rack.

    As others have stated.. 48v is a long standing standard for telecom power.

    • by Skapare (16644)

      But 48VDC also means dual conversion. Convert the AC to 48VDC, then do the conversion again with the PSU in each chassis. You have to get both conversions to be very, very efficient to make that worthwhile.

      Everything from Cisco can be had with 240VAC. Very little telco equipment these days actually requires a 48VDC power source. And most of that is for telcos, not for web site providers (for example). And where big network providers do need some 48VDC-only equipment, that can usually be put in the nort

      • by PPH (736903)

        But 48VDC also means dual conversion.

        Not really. Every switched mode power supply converts AC to DC, then back to AC (at a very high frequency) and then back to DC (at several voltages). The whole DC buss distribution idea pulls the first AC to DC conversion out of every individual supply and centralizes them. This makes it possible to back up the DC buss with batteries. But as others have noted, the high fault energies available on these busses are harder to deal with using common circuit breakers.

      • by tlhIngan (30335)

        But 48VDC also means dual conversion. Convert the AC to 48VDC, then do the conversion again with the PSU in each chassis. You have to get both conversions to be very, very efficient to make that worthwhile.

        The problem is, you need high voltages. You cannot run 12VDC to every server because you're talking about HUGE currents.

        Let's say the server is high powered and takes say, 480W. At 120VAC, that's 4A, maybe 5A after power supply inefficiency. 5A isn't a lot of current and wires are nice and thin (like they

  • by Skapare (16644) on Wednesday March 14, 2012 @05:08PM (#39357753) Homepage

    ... convert that AC to DC at a "blade rack". That would be a rack designed to take blades. But the blades would be a mix of

    • Processor blades (mostly)
    • Power conversion blades
    • Battery backup blades

    This will safely segment the power, leaving the DC busses limited to the amperage needed for one rack ... or even partial rack. It also has the flexibility of balancing power conversion vs. 1st tier power backup (at the point of use). Increasing the backup times to a couple minutes allows slow start generators, which are more reliable.

    I would run 416/240 three phase everywhere in the data center (even in North America ... transformers for this are readily available). Where equipment isn't on the DC system, run it on 240VLN. The AC/DC converters might run on 240VLN or 416VLL. In countries with 400/230 or 380/220, just use it that way direct.

    AC is safer due to the zero crossing. Circuit breakers can break a lot more power (usually 5x the voltage) with the advantage of AC, as compared to DC. A 380VDC breaker for a rack would be HUGE, especially if it has to handle a data center level of fault current.

  • what about PSU with build in UPS hook ups? so you can get rid of the AC to DC to AC to DC part and make it just AC to DC? at each system?

    • by Skapare (16644)

      Are you talking about a 2nd AC input, or a separate DC input which can be supplied direct from battery?

      • by nschubach (922175)

        I assume a DC connection that you can plug into a bank of batteries that can be used for power if the AC should fail (and charging the batteries when the AC is on).

  • Verari Systems (Score:5, Interesting)

    by sdguero (1112795) on Wednesday March 14, 2012 @05:49PM (#39358165)
    I worked there for 7 years. I'm not going to get into specifics but I will say:

    Verari tried to take advantage of the efficiency gains in DC with exotic power supplies etc... And that company went the way of the dodo bird after trying to force 800V, 48V, and 12V DC power distribution systems in customer data centers. The fact is, everything already out there (switches, routers, servers, etc) uses AC-DC power supplies in each unit and it works in 99% of power outlets with pretty good uptime. The added complexity of running DC infrastructure isn't worth the efficiency gains (which on paper sound like a lot but theory rarely translates to reality the way we think it will), and when one DC rectifier burns up and takes down a hundred servers (vs 1 server with an AC-DC supply), customers aren't happy. Between the uptime issues and employee safety concerns (high amperage DC power is more dangerous than AC for a variety of reasons) it's also a liability nightmare

    Again, I don't feel like getting into specifics but modern datacenters != underground telco installations and DC power distribution has a LOT of challenges that are often overlooked when marketing types start squawking about efficiency gains.
  • by 1sockchuck (826398) on Wednesday March 14, 2012 @07:07PM (#39358843) Homepage
    The effort to gain acceptance for DC distribution in data centers is being helped by a series of investments by ABB [datacenterknowledge.com], and the growth of the EMerge Alliance [datacenterknowledge.com], which is trying to unify DC proponents around a 380V standard. The challenge for DC is that customers don't ask for it, meaning multi-tenant facilities aren't likely to offer it. Also, Schneider says it is "not aware of any data centers moving off of their established, traditional power distribution to DC." In fact, NTT has at least five DC data centers in Japan, and ABB is backing a DC distribution project at a Swiss hosting company [datacenterknowledge.com]. In the US, there are numerous sites testing DC power, which is widely used in telecom infrastructure.
  • Is it just me, or do these 'DC in the data centre' articles seem to get posted to /. every few months?
  • Telecom already acknowledges DC as the victor. It's about time the datacenter people also recognize the efficiencies of DC power in the datacenter.

  • Electrocute an elephant with AC: http://en.wikipedia.org/wiki/Topsy_(elephant) [wikipedia.org]

"Go to Heaven for the climate, Hell for the company." -- Mark Twain

Working...