Slashdot Mirror
Was Thomas Edison Right about DC Power?
Declan McCullagh writes "Everyone knows the alternating vs. direct current wars ended with Thomas Edison and Nikola Tesla. But now DC power is being seriously considered for data centers. DC advocates say that plugging servers into AC power is inefficient, and switching to DC cuts down on waste heat and component failure. The University of Florida has even bought 200 DC servers."
In Washington State, Verizon (Was once GTE) runs almost all DC powered servers and Telco equipment in their Data Centers. Many of the IBM server my company buys support DC power.
For moving power over long distances, AC is king.
Nope. For the longest-distance transmission lines, you see DC being used. There comes a point when the capacitive losses you get from using AC encourage you to switch to DC, and for lines of several hundred miles, you start seeing DC transmission lines.
for crap's sake, dc powered servers are nothing new, many have config option of "-48VDC standard telco" supply.
In addition to capacitive losses, there is also the fact that you have to dimension your transmission lines to handle up to Vp (peak), not just Vrms which is what controls the amount of power that actually travels through your system. In effect, by going to DC, you can run the whole system at 1.4 times the voltage, and run more power through the same wires with no additional losses (other than conversion.)
That said, if space and cooling are an issue it might well make engineering sense to get the transformers, capacitors, and rectifiers out of the computer boxes. Big 5v/12v power busses wouldn't even need to be insulated. So while the reporter badly mangled the story, the engineering sounds reasonable to me.
Actually, that's the norm across the phone industry. Everything, and I mean everything, runs on -48V DC. Okay, not the fluorescent lights....
This goes back to the telephone talk battery, which is -48 V DC. That powered the phones via old cord switchboards, and was the voltage of electromechanical (stepper, and later crossbar) switches, which basically used relays. Electronic gear was then designed to run on the same power plant. A telephone building has a big bank of batteries, powered by multiple "rectifiers" (DC supplies) which, btw, are normally engineered to not run over 40% of load. (That way they can still run the systems and recharge the batteries when one of them is kaput.)
If you then put anything else into one of their buildings, the Network Equipment Building Standards (NEBS), which are Telcordia documents that practically carry the force of law, dictate that equipment be DC powered. Among other things -- NEBS gear has to meet the brick schytthaus test. (Sun Netras and many Cisco routers meet NEBS. Your basic rack server doesn't. And aluminum racks are STRICTLY forbidden; it has to be steel.)
So because of the talk voltage on analog phones, lots of computing equipment is engineered for -48 V DC power. Sort of like the legend (I know, that one is not really true) about the railroad track gauge being based on Roman chariots. But in this case it's surprisingly effective.
The article mentions the distribution will be done using 48V for distribution - you will still need DC:DC conversion at the boxes. These DC:DC converters will need to be run at higher current than an AC:DC converter. Higher current can cause more series loss in the system as well leakage losses in the switching supply.
AC:DC converters, as mentioned in the article aren't really that ineffiecient (article itself quotes 90%). AC:DC converters are infact really DC:DC convertors, they just have a rectifier circuit to convert the AC to high volatge DC for DC:DC conversion.
The origin of the 48 volt number is that it was convenient, and now it just sneaks under the 50-volt "low voltage" cutoff in the NEC, which I think was written with telcos in mind. The glorious thing about this is that you don't need licensed electricians to do power wiring in a central office.
And the reason it's negative with respect to ground goes all the way back to the telegraph system: Western Union initially ran bipolar lines and noticed that the positive ones corroded much faster. Sodium ions (from dissolved salt) are negative, and thus repelled from lines that're also negative. The whole phone system was built with positive ground because of this, and it's saved incalculable maintenance costs. It does tend to mess with people's heads the first time, if they're used to negative ground systems, but you get over it quickly. (A number of traditions use blue for "hot" and black for ground/return, to help escape your "red equals positive" association.)
DC power as used by telcos is also always redundant. There's an A-side and a B-side for everything, and the cables are sized so that the entire load can run from just one side. This leads to some very fat copper, which is cheap compared to downtime. You don't achieve five-nines reliability with a system that contains single points of failure!
Now, about rack-mounting: This was also invented by the telcos, originally in a very wide (40-inch?) format, for the panelboards and Strowger switches. Some of the old crossbar equipment is still in those huge racks, but the 23-inch width is infinitely more common now. All telco equipment is mid-mounted, with the ears approximately in the center of gravity on the shelf, so the force on the screws is shear. There's no torsion on the mounting flange unless you step on the front or back of the shelf. Cooling is always convective bottom-to-top, or occasionally front-to-back with fans. This leads to a "cool" front aisle and a "warm" back aisle between alternating rows of equipment.
Now, the pro audio industry borrowed the rackmount idea fairly early on, but they were mostly mounting control panels and mixers, which are very shallow, so flush-mounting made sense. They also changed the every-inch Western Electric mounting holes to an alternating-spaces "EIA" standard, and narrowed the rack from 23 to 19 inches.
Somewhere along the line, an absolute idiot decided that computers should be rackmounted, but they should be 19 inches wide, flush-mounted, and use EIA hole patterns. I'm sure this has something to do with mainframe legacy getting perverted by peecee people. The current mishmosh of mounting standards (19" vs 23", two-post versus four-post, flush versus mid, inch versus RU, front-cable versus rear-cable) is what every datacenter tech deals with on a daily basis. Throw overhead racks versus raised-floor cabling into the mix, and you've got a recipe for frustration!
If you're familiar with the concept of "blade servers", where common components are separate from processor resources in the shelf, congratulations. Telco hardware has been built like this since the invention of the circuit board. Actually, the concept of replacable plug-in units goes back before that, but it got vastly easier with printed wiring boards and card-edge connectors in the sixties. Most of the "good ideas" in serious computing circles are actually century-old ideas in the telco industry. Spend a week shadowing a central office tech before you design a datacenter, please!
Also consider: If your datacenter is already built for DC, throw some solar photovoltaic panels on the roof. Inverters are a large part of most PV systems' expense, and you can skip that part. Why not start offsetting your grid demand now?
Also also: Edison was flat-out wrong about DC. The modern switching power supplies that make DC transmission lines practical didn't exist in his day. Besides, long-distance power transmission is an entirely other discussion.
Tesla believed that electricity should be free, so he created a tower that transmitted electricity over a distance. http://en.wikipedia.org/wiki/Wardenclyffe_Tower
The article conflates several things.
First off: Digital electronics generally requires several voltages. And they're all low, requiring high currents, massive conductors, and local filtering and regulation. So even if you're providing DC power from outside the room, you'll have a switching power supply (or several) in each piece of equipment to convert whatever the rough DC power is to whatever you need, smooth it, and regulate it.
But while some electronic devices use a common switcher to generate all the voltages with one conversion step, others use a "roughing" supply and a bunch of local supplies. Part of that is to get better regulation - part is because the roughing supply must run from 60 (or 50 or whatever) Hz and thus requires big caps to tide you over the low part of the cycles - caps you don't want taking up space near the components.
If you're going to do it in two stages anyhow, you can put your roughing supply OUTSIDE the room and only have the final supplies inside. The roughing supply has a lot of heat dissipation so you save a bunch on your cooling.
Second: There are two standards for power distribution in electronics rooms:
- Your local power line stuff. (120/240/480/208-3-phase in the US)
- The telco standard: x2-redunant 48V DC.
A lot of equipment - especially networking equipment - is manufactured for sale to tellcos and other operations that use the standard. They might have initially used it because some of their equipment was co-located in tellco sites, where only 2x48VDC is available - and they got a quantity discount for buying a bunch of the same stuff and went to 48V for their own sites. Or they might use it because it's MUCH simpler to do backup power with floating batteries and century-old technology than with a building-sized UPS. (Note that a UPS CAUSES at least one outage when first installed and on the averate at least one more within the first year of operation from some malfunction. And a UPS dissipates more power than a roughing power supply or a battery charger.)
But the standard for 48VDC is REDUNDANT 48VDC supplies, with the equipment only requiring one (and typically doing "cutover" with diodes B-) ). With the equipment already set up for redundant supplies it's not a lot of cost or work to wire both sides and put in two 48V feeds to the equipment room. (Four diodes are a LOT cheaper than a pair of 120V roughing power supplies at each box, too.) So of course the users of such equipment normally give it dual supplies. (Even if it's a single rack and so they just put two roughing supplies in the rack fed from two different 120V feeds.)
The result is that all the equipment has redundant power supply, and keeps operating glitch-free through a number of kinds of partial outages - AND power supply repair and replacement. This is what's responsible for much of the claimed increase in reliability.
The whole Edison/Tesla DC/AC war had to do with the economics of CROSS-COUNTRY power transmission. AC beat DC there because a century or more ago it was virtually impossible to jack DC voltages up to levels suitable for long-distance transmission and back down to levels safe for distribution within houses, while AC could do that easily and efficiently. So Westinghouse/Tesla could ship cheap power from Niagra Falls to New York City while Edison had to build fuel-burning power plants IN the city. It has essentially nothing to do with shipping the power around within a single building.
Bantam Dominique roosters crow a four-note song. Once you've heard it as "Happy BIRTHday" you can't NOT hear it that way
Edison and Tesla were both right. Remember, the DC vs. AC wars were fought back when the load was mostly made up of lights, motors, very utilitarian things. AC is fantastic for transmission over long distances (and for running three phase motors, but that's another story). DC happens to be better at running precision equipment like computers -- heck, they all run on DC already. All we're really talking about here is taking advantage of an economy of scale by doing one big power supply (or a few, for redundancy) instead of one for each machine.
Ever seen a telco rack? Everything runs on -48VDC. Everything. A telco rack always includes a couple of DC power supplies, and all the equipment just ties in to a common DC bus. The best part of all: the UPS simply consists of four "car batteries" (not exactly, but you get the idea) wired in series and tied directly into the bus! No pesky inverters to deal with.
The telecom industry has been doing it this way for decades. It's about time the computer industry got on board.
Tired of FB/Google censorship? Visit UNCENSORED!
Really though it's not that good, it only gets 4.2 stars at imdb.
-s
Meanwhile back in the real world ...
In a standard PC power supply the incoming AC is rectified and stored in a capacitor. Energy only flows into the capacitor when the voltage after the rectifier exceeds that stored in the capacitor. This results in a waveform which departs considerably from a sine wave - no current flows for most of the time while much higher currents than expected flow at the peaks of the half cycles. Electricians interpret this as a bad "power factor" from their experience driving inductive loads where the current lags the voltage by as much as 90 degrees.
Standard PC power supplies are nothing like 90% efficient largely because of this crude rectification of the mains. Compare the rating of your supply in watts with the input voltage multiplied by the input current. These values should all be marked on the case.
Power Factor Corrected (PFC) supplies are available. The better ones use a switch mode circuit to charge the reservoir capacitor through most of the main power cycle, while the less good ones incorporate a capacitor across the mains to buffer the large peaks of current when the input voltage exceeds that stored in the reservoir capacitor.
One advantage of AC is the ease of transforming it to other voltages using transformers and the ease of using it to drive motors especially with multiple phases. In the modern age where switch mode power supplies are cheaper than those using transformers operating at mains frequency this advantage no longer exists. One disadvantage of using DC is the difficulty in switching the stuff off - inductance in the load drives the current straight through an opening switch or fuse creating a nice sustaining arc which is not quenched by the current dropping to zero twice each cycle.
Edison "advocated" for all power systems, including longdistance transmission, to be DC - because that's what he was selling. His battle with Tesla for the first big contract, electrifying NYC, is the stuff of legend. Tesla won. And died penniless in the 1940s, while Edison died fat and rich from thousands of patents, most on inventions invented by people working for him. Some of whom no doubt died penniless.
--
make install -not war
Power supplies that only draw current for a tiny part of the mains wave, to top up the capacitor, are banned in the EU because too many of them can and did affect upstream power equipment.
I am one of many. My idea is not unique, nor do I expect my voice alone to sway you. I speak in a chorus of opinion.
It turns out Edison was not completely wrong: HVDC
In particular, "Increased stability of power systems" is certainly something that individuals in the Northeastern US and London may be interested in.
Of course, AC still has its uses, but the chart is now thought to be:
really long distance -> HVDC
long distance -> AC
short distance -> DC
The only limit on using higher frequencies comes when you start to get magnetic losses in transformers and chokes, so in practise a few hundred kHz is the useful limit in switched mode PSUs.
Thus, if you were starting again with an electric grid system, 500 or 1000Hz would be a much better solution.
IAAPD (I am a PSU designer)
1MHz SMPS is nothing fancy these days. In fact they're available even as integrated chips which combine FET switches and the controller into one IC. If you count in point-of-load charge pumps and such, you see up to 3MHz.
Generally speaking, worst offender is high-current FET gate charge which eats up more power than all the other losses combined for synchronous buck transformer (higher DC to lower DC topology, most common type in use probably) Small (1-10uH) inductors are much better behaved in comparison. One reason to use such high frequencies is indeed that you can get smaller inductors and you have less ripple current.. But you're limited by the fet gate charge. Of course, if you're driving some 200W load, you can just say "h*ll with it" and build high power driver circuit to drive your switches, 5% waste heat on switching losses doesn't bring down the house when you can use dirt cheap inductors.
Cheap PSUs support it because they are banned in the EU now if they don't support it.
Most cheap ones use the less good method he referred to(Often called 'passive' vs. 'active' PFC, in PSU literature.)
Remember the death of Archimedes!
Anyway, the respondant claimed:
Check your history!
Edison died nearly penniless too.
The account which I read described how he ran across some iron-rich sand on a beach, and it gave him the idea to try a new mining technique where the ore would be extracted from non-ore material by dropping the sand past magnets. The idea was a good one (and is still used) but the site he chose to build his iron mine turned out to be almost completely lacking in iron ore. The iron ore in the sand on the beach had apparently washed up from some other source.
Maybe he wouldn't have been desperate enough to try such a risky thing if he had been ALLOWED to sell AC power. I'm sure he could see the advantages of AC for power transmission but Edison didn't have the patents for that, and you can bet that Westinghouse wasn't going to license the technology at a reasonable price to their chief competitor.
So Tesla got ripped off by Westinghouse because he wasn't business savvy and they got ownership of the patents. Then Edison, even though he was somewhat business savvy, got shut out by Westinghouse because they owned the patents. In both cases, patent law helped business-people who didn't invent anything get rich while the real inventors lost out. Shouldn't we remember that the patent system was set up in order to encourage invention?