Slashdot Mirror
DC Power Poised To Bring Savings To Datacenters
snydeq writes "InfoWorld's Logan Harbaugh follows up his '10 IT Power-Saving Myths Debunked' to argue in favor of using DC power in the datacenter. The practice — viewed as a somewhat crackpot means for reducing wasteful conversions in the datacenter just a few short years ago — has gained traction to the point where server vendors such as HP, IBM, and Sun are making DC power supplies available in their server wares. Meanwhile, Panduit and other companies are working to bring down another barrier for DC to the datacenter: a standardized 400-VDC connector and cabling solution. And with GE working to list 600-VDC circuit breakers with the Underwriters Labs, DC's promise of reduced conversion waste could soon be commonly realized."
Tesla smiles in his grave as Franklin catches on fire from Nikolai's coil-arcs-of-doom.
'We are trying to prove ourselves wrong as quickly as possible, because only in that way can we find progress.' RPF
eat my asshole.
Who would have thought the GE would be a big supporter of DC.
If something is so important that you feel the need to post it on the internet... It probably isn't that important.
I felt a great disturbance in the Force, as if millions of Tesla fanboys suddenly cried out in terror and were suddenly silenced. I fear something terrible has happened.
Suggestion for the DC power supply designers: have a heart and build GFCI into the spec.
One of the important factors that was overlooked other than the inefficiency of DC over large distances is the risk of electric shock. DC is unforgiving and anyone who receives a shock at the higher voltage levels will have very little to no chance of survival as DC current polarizes the blood and there is no way to reverse that effect in time to save that person.
See the following for a basic description of what this is about.
http://www.cdc.gov/eLCOSH/docs/d0500/d000543/section2.html
http://hyperphysics.phy-astr.gsu.edu/hbase/electric/shock.html
The difference between direct and alternating voltages and currents is that one swings negative to positive thereby reversing the polarity of the potential while with dc everything is a constant supply and thus more harmful.
=Smidge=
Is it just my observation, or is eldavojohn an idiot?
rolls over in his grave.
Obligatory blog plug: http://www.caseybanner.ca/
What's the quote? "One Test is Worth 1,000 Expert Opinions".
So build a few variations and lets see what the deal is.
Nothing in the world is more dangerous than sincere ignorance and conscientious stupidity.
I don't run a datacenter, but I sure would like to get rid of the power bricks that all small electronic appliances seem to come with these days!
"[Regarding the 'cloud,'] ownership was what made America different than Russia." -- Woz
I know this is off-topic but please forgive that. I urge you to give someone a warm embrace and tell them that you love them, and mean it. There is not enough of this in the world today.
In the 80's we built custom interfaces for large computers using wire-wrap Standard Logic Inc. wiring modules. The planes of wiring were assembled into rackmount chassis which were fed DC power via a vertical bus-bar system in the rack. The busbars were about .5 X 1 inch solid copper, insulated by shrink tubing with holes cut for the threaded holes in the busbar. The power supplies were rackmount 100 or 200 A Lambda supplies providing either 5 volts or 12 Volts. It was occasionally a pain to be called into the computer center in the middle of the night to replace one of those heavy power supplies - at least they were at the bottom of the rack.
OZ
enough is too much
Switching power supplies have gotten much more efficient in the past few years. Now it makes sense for a standard DC bus to run everything. The telecoms have been doing this for ages.
Only the State obtains its revenue by coercion. - Murray Rothbard
I thought the power in D.C. caused waste and ineffeciency.
If brevity is the soul of wit, then how does one explain Twitter?
Telco gear tends to be 48VDC all over the place. It just works. Speaking as a guy working at a telco in the IT department, I'm hugely in favor of moving to 48VDC servers.
I'm not an EE. But back during the dotboom I thought it would make sense to have a big ups in the data center that output voltages that mother boards expected as input. I almost thought of rigging my own experiment using laptops as servers and feeding them all 12vdc directly from the UPS battery pack.
Ok rip it apart guys, why is wrong with that plan?
Think Deeply.
Pros:
No power supply needed for each machine. This removes a major point of failure. Instead, one would need to just step down voltages to the 5 and 12 volt rails. This also helps with cooling because the room AC/DC converter can be cooled with a dedicated system, either liquid, or part of the HVAC system.
Cons:
48 VDC needs a dedicated connector with a high plug/unplug cycle rating that people know is 48 volts and 48 volts only. It sucks when you have to manually wire it up, because this takes time and there is always the risk of getting zapped if you don't throw the right circuit breaker (or pull the right fuse) on a telco rack where 48V is in use.
Because there is only one 48VDC power supply for a room, it has to be held up to a lot more rigorous standards than average mains current. It has to not just provide 48VDC, but provide it under extremely heavy load without the voltage dropping by much.
Maybe 48 volts would be a new computer standard. The key is not having to wire it up manually like some stereo speakers, but giving it a dedicated, foolproof, power connector that Joe Twelvepack who is slurping down his seventh can of Bud Light can easily and reliably plug and unplug while staggering around in the back of the server room until his shift ends.
One can't help but reflect upon these two and their stubborn support of DC and AC respectively. Edison created a circus atmosphere demonstrating the dangers of AC. He electrocuted dogs & other animals and even participated in the design of the electric chair to prove his point.
Edison's financial ambition was part of the problem, and his inability to understand AC, but mostly it seems to have been an emotional attachment to DC.
Let's hope that in our time emotion and personal gain have no part in such decisions.
...omphaloskepsis often...
Alternate view: http://cim.pennnet.com/display_article/347089/27/ARTCL/none/none/1/A-powerful-debate:-AC-vs-DC-distribution/
Or, to summarize - if you take a high-efficiency AC system and convert it to 480 volts, downstep to only 240 volts (and all todays' boxes can run either 110 or 220-240), you can get to within 1% of the DC system.
Add to that the savings in materials (1.5" copper wiring? Booster cables for diesels aren't anywhere near that thickness) and there's no real reason to change.
In fact, the biggest saving would probably be if we went from 120v to 240v for everything. One less down-conversion, etc.
Kevin Smith on Prince
a standardized 400-VDC connector and cabling solution
I set this kind of system up myself and it works great, assuming you need a lot of cores. I strung together 296 Intel Core 2 Duo chips in series accross the 400VDC supply, so each one gets the specified 1.35 volts. If I want to overclock, I just take a set of alligator clips and shunt across a few dozen of the chips, and it boosts the voltage to the remaining CPUs.
The only problem is that with so many chips, I get occasional failures, just like I do with my old Christmas lights. Then I have to try shunting around each of the CPUs by trial and error until I isolate the burnt out one before I can get my cluster running again. Oh yeah, I also have to be really careful to keep any peripherals I plug in away from each other and/or grounded objects.
The only issue I really see with this would be a safety issue. 110VAC and 220VAC aren't particularly safe (anything over 50V AC or DC isn't safe to touch), but 600VDC is very scary to be working around. Aside from that I have no problems with the logic in TFA, besides a few minor errors: stepping down 16kVAC to 440VAC to 220VAC to 110VAC should only incur a total loss of less than 2% (more likely less than 1%) since transformers are typically efficient in the 99% and higher range.
Just a side note, this has already been growing in the field of UPS units for at least 5 years, and it's not terribly hard to find UPS units and PSU units with DC connectors.
(Since to use a UPS without DC means converting battery's DC, sending it to the PSU in AC where it's converted back again.)
In Soviet Russia jokes are formulaic and decidedly non-humorous.
as in all those datacenters saving power through dc, actually receiving the electricity from the national distributing network through ac ?
Read radical news here
It's great news when a brain-shake idea that started off as a fly in the wall is able to stand up and be back in business with a flick of a switch. When dealing with such high voltage and when dealing with rising power costs from power companies who got you by the balls, ideas like this shake your foundation. It's time to show business you ain't got hold on me. It isn't important who made who. Let's make it. The IT budget must get back in black.
And I say it with a stiff upper lip.
P.S.: AC/DC fans, I salute you.
P.S.2: cowardly posted to protect the guilty.
From your post:
... that Joe Twelvepack who is slurping down his seventh can of Bud Light can ...
From another TFS:
... helping Joe Six Pack fight aging and cancer with every swill of beer ...
There's a correlation somewhere here, but I can't figure it out.
Comment removed based on user account deletion
> I don't run a datacenter, but I sure would like to get rid of the power bricks...
DC vs AC wouldn't help you rid yourself of power bricks. No more than it can help a datacenter get rid of power supplies in each server. Telco equipment runs on 48 volts not to save electricity but because of the way telephone exchanges are built. Telephones don't go down, period. So how do they accomplish this miracle? Huge battery banks. Back in the day a DC-AC conversion system large enough to run a whole switch plus drive every telephone would have been all but impossible. So they just ran everything directly from the batteries and used the mains to charge the batteries.
This DC in the datacenter thing is just a green craze that will pass. It is pure unadulterated snake oil. Go reread the summary. They ain't even doing the smart thing and adopting the telco 48V standard. Does anything in a server run on 48V? No. Does anything in a server run on the 400V they are proposing? No. So a DC-DC conversion will be needed, i.e. a switch mode power supply. Guess what is in a current server? A switch mode power supply. Current PC power supplies are available with efficiencies over 90% without buying too far off the mainstream. I seriously doubt these DC powered supplies will be much better and in the end that is the ONLY number that matters. Except these DC installations have to factor in the power loss from the big AC-DC conversion and worry about redundency, backup power, etc.
Democrat delenda est
That concluded that using the european system of 230/400 3 phase AC for distribution splitting out to 230V single phase AC near the point of use was almost as efficiant as a 400V DC system and far cheaper and easier to deploy. Your servers existing power supplies can almost certinaly handle 230V without any problems (changing a switch may be required on crappier models)
BTW in many cases there are often huge savings to be made without changing your infrastructure just by using better PSUs, cheapasss PSUs are both inefficiant and unreliable.
note: i'm known as plugwash most places but i screwd up registering that here somehow in the past and now can't register
Parent is a troll, stop modding him informative you idtios.
-Tesla
There are a number of companies providing commercial DC solutions for data centers. Validus DC Power is providing products for DC power distribution, while Power Loft is building a brand new data center optimized for DC power.
RichM
Data Center Knowledge
These days they are usually switching power supplies, which are quite efficient (not to mention smaller).
From TFA:
The power starts at the utility pad at 16,000 VAC (volts alternating current), then converted to 440 VAC, to 220 VAC, then to 110 VAC before it reaches the UPSes feeding each server rack.
That's just stupid. I hope it's just a case of a journalist not correctly understanding (which is a common problem). Given the usage of numbers like 220 and 110, instead of the standard 240 and 120, I do suspect it is a journalist giving wrong info. But even many computer people don't know what the standard power voltages are (and have been for decades). Lots of people in the USA still refer (incorrectly) to "two twenty" and "one ten". The standard in Europe is 230 volts.
With so many conversions taking place, there will be a lot of power loss. To begin with, the computers should have been operated directly on the 240 VAC, not 120. That 240 VAC should have been obtained from the utility power directly (though voltages like 7200, 7620, 7970, 12470, 13200, 13800, 14400, 19920, 22860, 23900, 24940, 34500, etc, are more common ... I've never heard of 16000 being used). Since power comes in as three phase, the ideal voltage conversion would have been 240 VAC line-to-neutral, which would give 416 VAC line-to-line. Neutral harmonics issues can be avoided by use of oversized neutrals or multiwire neutral.
Do AC wiring correctly, and the advantages of DC are minimal at best. Where the DC plan can have an advantage is that the conversion to 400 VDC, done on a large scale, can be done more efficiently. If that doesn't happen, then it's just one AC-to-DC conversion vs. another AC-to-DC conversion. When the 400 VDC gets to the computers, you still need a PSU to convert the 400 VDC to the various voltages provided to the components inside the computer box (e.g. 12V, 5V, 3.3V, etc).
AC voltage conversion can be more efficient than 98% when properly designed low impedance transformers are used. That can beat the DC conversions ... even DC-to-DC, in most cases. So you want to do conversion of DC only once or certainly no more than twice.
It has been reported that mainboards can be designed to efficiently convert 12 VDC to the other voltages needed. Google's original proposal was to supply computers with 12 VDC, allowing them to be manufactured without the PSU entirely, and thus in a smaller footprint as well as having the increased efficiency. The 12 VDC would come from a large PSU in the middle of the rack (to limit the length of wire carrying the higher current that is involved with a low voltage). That large PSU would be designed to accept AC at any voltage from 380 to 480, 50 or 60 Hz, and thus be usable just about everywhere in the world. The PSU may even operate more efficiently when fed with full three phase power (the full cycle nature of three phase power reduces the level of filtering needed for smooth DC).
Running DC is NOT a crackpot idea. It just needs to be studied correctly, in its various possible forms, and compared to CORRECT designs of AC wiring, in its various possible forms. The choice of 400VDC for distribution within a data center to the individual PSUs is a reasonable one, given that the existing PSU designs go through a conversion to 340VDC to 380VDC, anyway. But these same PSUs, especially in the larger form of one per rack, could just as well be designed to operate from 380 VAC, 400 VAC, 416 VAC, or 480 VAC.
Maybe DC is the right choice. Or maybe AC can still be the right choice when engineered correctly (which far too often is not done, sometimes due to ignorance, sometimes due to budget limitations which would never go for DC anyway, and sometimes just due to mental inertia).
now we need to go OSS in diesel cars
The reason Tesla/Westinghouse won the current wars with Edison because there wasn't any good way to step DC voltages up or down. You can't transmit power very far at 110 volts. AC allowed the use of inexpensive and transformers to step voltage down at the customer site and transmit at high voltage over long distance.
Today solid state converters do allow the step up / down of DC voltage, and very high voltage DC can be sent over long distances with less loss than the same AC at the same voltage. At least one power company is looking at using DC transmission lines over long distance.
AC power still makes more sense for consumer and most industrial use, but for transmission and delivery of power in bulk DC seems to be making a comeback.
Fewer. It takes fewer wires. Or it takes less wire.
This is nice, but I see it as interim until we can produce superconducting puddles at room temperature.
Moving the AC->600VDC stage out of the controlled environment will be a savings even if you keep the inverter and stay AC in the datacenter.
for actually going DC in the datacenter, the top benefit is losing the inefficiency and heat of the inverter stage of the UPS.. Instead, you have the potentially smaller losses of several smaller 600VDC to 48VDC converters in the racks and potentially cheaper power supplies that don't have to care about power factor.
The con side is the need to re-fit, heavier power cables from down-converter to the individual machines and the underfloor area becomes much more hazardous (600VDC = 3rd rail).
I work in the telecommunications industry. It has always been standard practice (at least where I work) to use DC power supplies for data equipment if they are co-located with voice equipment, since most voice equipment uses -48 V DC power.
This has the additional advantage of utilizing the battery backup system (required for voice) to also back up the data equipment's power.
I have a bad feeling about this...
I remember previous /. discussions in which suggesting DC in the datacenter got me accused of crackpottery.
It also wasn't so long ago that Linux and OSS were viewed as crackpot ideas. Today I was interviewed for two jobs: one as a SuSE Linux Administrator, and the other as a Solaris Admin. They offered identical salaries.
"I assumed blithely that there were no elves out there in the darkness"
Half as many amps == half as many power strips, half as many UPS devices, half as much wire, etc.
In the split-single-phase arrangement that is used in the USA, the only difference is whether there's a neutral wire in the conduit. For a given wire gauge you don't get any more power from a 240V circuit, because they're fundamentally the same thing, just one has kind of a "center tap". That copper is a very marginal savings (3 conductors vs 4) when you figure all the labor, conduit, breakers etc that's going to be put in anyway. And if you're dealing with 3-phase it's even less (4 vs 5 conductors).
In a colo environment it would be smarter to run 120 (with shared neutral) so people can use the normal plugs and cables that they have on hand, although in a single-customer datacenter where all your equipment is sure to have modern power supplies, fine, go with 240. But it's not hard to wire 240V outlets as needed (eg for a high density unit like a blade chassis or cisco gear).
You don't use any fewer power strips because you still need a plug per computer regardless of the voltage, and you still need to same amount of UPS equipment because your VA and WH would be the same for a 120 vs 240v UPS of a given price or physical size. It may surprise you that 120V and 240V UPSes generally have the same internals, the only difference is the plugs and cables that they're outfitted with on the back panel. Try measuring the voltage across two hots on different plugs of a "120V" UPS - you'll probably see 240V.
... That thought it was going to be something about an upcoming Batman or Superman (or some other DC Comic franchise) comic where they fight terrorist that were poised to blowup major datacenters?
I Guess so.
You might be surprised to know that the very people who have identified DC power infrastructure as a useful thing just happen to be selling products to help you implement it!
...checks portfolio... ...finds owner of Duracell... ...buys more Proctor&Gamble stock...
Vision with execution is hallucination.
I've seen a joke go over a person's head, but not three heads in a row.
The responses to this here where highly predictible, and many af them are quite naive.
Modern DC-DC converters have excellent Efficiency over a wide dynamic range of loads. This holds true for the small, nice isolating ones which every designer of instruments likes very much, and also for larger ones. No transformers, smaller capacitors, easier redundant designs, easier buffering. In a time when computers are more and more designed to vary their input power according to their load, all these things could provide a savings of energy (and money). Even if this saves only a few percent, the investment will be payed off in a reasonable time.
I just one want one plug into the wall outlet and several DC connections to the array of external WD MyBook drives instead of having each of them plugged into an outlet of a power squid.
Oh, say does that Star-Spangled Banner entwine / The myrtle of Venus with Bacchus's vine?
For those of you not using Firefox and NoScript, InfoWorld's website pops an error with Firefox that claims InfoWorld would like to run Javascript marked as UNSAFE. That is, it will have access to all windows and panes, which is plain evil. Tread carefully please, especially if you are using Internet Explorer. This, of course, assumes that you will RTFA.
The AC vs. DC discussion has focused on efficiency, but I'd think reliability would also increase with DC distribution. I've found the power supply to consistently be the most unreliable component in rackmount servers or blades, usually simply failing but at least once catching on fire. Some servers have redundant power supplies, but I'd rather have the redundancy at the rack level rather than at the server level. VDC converters would presumably be higher quality components as well, at least initially, until manufacturers start cutting corners just like they do on existing AC/DC converters.
I can't believe no one has referenced Google yet... I read a paper a year or two back on how they were making their data centers more efficient. Now, lo and behold, they have a website dedicated to how to run an efficient data center - surprise!
Check out
Instead, we're rehashing the old arguments between Tesla and Edison??? George Westinghouse backed the winning idea AT THE TIME. Get over it, people...
They seem to claim savings from two things:
- Raising distribution voltage to 575 V from 48, 110, or 220 V, minimizing distribution losses.
- Consolidating conversion apparatus into one place, minimizing conversion losses.
The first point is valid, although it doesn't mandate DC per se.
The second point, however, needs a closer look. AC step-up and step-down are reasonably efficient, as well as rectification and filtering of AC. However, there are still DC/DC conversions from 575 to 48 V at the rack and from 48 V to the final operating voltage(s) at the equipment. The 575 to 48 V conversion is said to be 95% efficient. (In other words, it wastes 1 watt for every 20 watts converted. Worse than a good AC distribution transformer.)
At 95% for the DC/DC conversions and 97% for the AC/DC conversion, we get 13 kV -> 575 V -> 48 V -> device internal = 0.97*0.95*0.95 = 87%. With 99% efficient transformers and 90% efficient device power supply, going from 13 kV -> 440 V -> 220 V -> device internal = 0.99 * 0.99 * 0.90 = 89%.
Even if all the DC/DC conversion in the proposed scheme were 99% efficient, the maximum savings in power consumption would be in the order of 9 %, not 50 %.
Note that I've used the American two-phase system as an example here, which means there's no conversion from 220 to 110, just using one of two phases. Here I think the article makes an error, since it mentions 220 to 110 as a conversion. In most of Europe the incoming (or possibly UPS-generated) supply would be three-phase, 400 V between phases, 230 V phase to neutral. This would be brought to the rack and distributed as three single-phase supplies to individual devices, or possibly as one three-phase supply to a single device with great power demand. Up to 86 kW can be supplied with a single plug-in cable.
Usage: km/h for speed (kilometers per hour); kph for very slow impulses (kilopond hours).
Does that mean they'll have to figure out ways to reduce the noise generated by arc-fault interruptors? I know that residential AC AFCI breakers have an obnoxious acoustic hum, but I don't know if that'd translate into powerline noise for DC versions..
The server cluster at work is only like half the size of a full grown elephant, so how can I trust AC power not to shock anybody using the rack's KMM to death?
Additionally, comparing cost of energy lost in wiring per year...
365 days * 24 hours = 8760 hours per year.
Price of energy at $0.10/kWh.
0.845kW * 8760hours * 0.10 = $ 740 per year for 10A
3.300kW * 8760hours * 0.10 = $2890 per year for 50A
So for a large data centre its not about the ongoing cost of power losses in the wiring, its more the installation cost and minimising transformer power losses.
I know this article is primarily concern with Data Centers, but maybe we have come across another use for these 48 V (or something similar) DC sources(I see that some are skeptical of even doing the DC thing, because it may be more efficient to move to 240V), but if we did decide to create a DC infrastructure to make conversion to DC more efficient, then maybe we should think about making a widely compatible DC infrastructure (as we get $$ value from economies of scale). My thoughts are in relation to future Electric cars. I know that most will initially be designed to be charged via 120V AC which is widely available (especially in homes), but think about car "charging stations" which might be used at offices or other common places (Restaurants ?) where there might be the need to provide 48 Volts to a large number of electric cars at the same time(or we will have inefficiencies of millions of AC/DC conversions for all the new cars) for charging. Maybe an infrastructure could be built to handle both needs (datacenters and future electric cars) if we thought about things ahead of time and planned for this(initially the cars could be built to charge off of 120 VAC or something like 48V DC. Just a thought.
If you set the switch to 220 on your power supply, it should run fine on 320VDC. If the diodes over heat and short out, it will stop working on AC, but keep working on DC. It won't work with the switch set to 110 because that activates a diode capacitor voltage doubler that bumps the 120VAC up to 300 - 340 VDC internally and it needs AC to work.
Pros:
* Greater efficiency as the power factor is corrected in one central efficient AC-DC converter.
* No power factor correction is needed in each power supply
* Slightly less power loss from not going through two diodes in each power supply (0.7 or 1.4 volt dropout eliminated), and no power correction circuitry either if it has it.
* 220VAC power supplies may already work on 300 - 340 VDC without modification.
* You can hook 32 12.5V batteries up and you have a 400VDC UPS with no efficiency loss.
Cons:
* Bigger circuit breakers that can switch DC are needed (a 120VAC switch can switch 30VDC at the same amperage usually)
* Possible danger of arcing at 400VDC.
* Some new equipment needed.
I think computer makers like the idea of selling a bunch of new hardware while "being green". I think it should be made to be 320VDC not 400VDC