Slashdot Mirror
Data Centers And DC Power
mstansberry writes "In the final article in a series on the price of power in the data center, IT pros weigh the pros and cons of direct current-powered servers. A limited number of companies make servers with the power supplies removed with DC power distributed to multiple machines from a single unit. It saves power by skipping an extra conversion from alternating current (AC). Telcos have been using this method for years, but some data center pros are leery of taking on the new systems. It's not something people are familiar with and if they break down, you have to hire a specialized engineer to come fix them. But if they're saving even half of what they're reported to save on the electric bill, companies could afford to hire the engineers." We've reported on previous articles in the series.
Tesla, you're fired. --Thomas Edison
try { do() || do_not(); } catch (JediException err) { yoda(err); }
I've always wondered (from a non-technical point of view) whether there was a benefit in having our homes wired up with two sockets (or maybe a 5 pin mains plug) giving standard AC voltage and a low-current DC voltage as well (12V?). So many devices only need low voltage, wouldn't we all benefit in having a power system in our houses in this way?
Jolyon
Please read my Canon EOS tech blog at http://www.everyothershot.com
ac goes into data centres, systems run on dc. Either it gets distributed to each computer as ac and converted in a medium-sized box in the back of each system, or it gets converted in one big box and distributed to the systems as dc.
The question is of the efficiency saving of doing all the converting in a big box against the efficiency loss of piping it around the data centre as dc, and wether you get a large total net saving (which I suspect that you do, since even inside the data centre, it's not going far)
FGD 135
trouble shooting and correcting DC power is simpler than working with linear power supplies. Unfamiliarity is the problem, not the technology.
Any truly serious data centre would already have at least several power engineers on their staff.
If such a data centre is just now considering bringing such people in, then they have serious operational problems. They're not getting professionals in to do the jobs that professionals must do.
Cyric Zndovzny at your service.
Nope, you save two conversions.
Without DC distribution, you have AC->DC->AC in the central UPS, and then AC->DC in each computer's power supply.
With DC distribution, you have AC->DC in the central UPS, and no conversion in the computers.
You get down from 3 conversions to 1.
I started working with IP in a small ISP. We were bought by a loal Telco and over the years have got used to having all our routers and switches running on DC current.
One thing telco companies do well is DC power, they have alot of skill in providing multiple DC feeds from DC power systems, with battery backup and generators all in line.
I would imagine that any big server farm would benefit from this kind of setup. Especially when you have people runnnig the lines that are as good as some of the guys in the telo world, they can really make the wiring look like a art in some places.
...this may or may not save a step.
:)
However, it does provide a few significant advantages.
Telcos use DC because it's easy to battery-back. Since all your gear is already running from the DC supply, there's no guesswork about whether your UPS will be able to handle the load. Each piece includes its own converters, so all you have to do is size the battery bank. Since most telcos aim for 8-hour runtimes on battery (long enough to discover and fix a generator problem), overkill is the order of the day.
There's also the point that you can run several small generators, instead of one large one. In an AC world, keeping multiple generators syncrhonized is nearly impossible on a small scale, so you just run one big one. If your setup grows, you rip out the old generator and replace it with a larger one. In DC, since all your generators feed the same battery bank, you can just tack on more capacity without trashing your original investment.
Using multiple generators provides cheaper redundancy too. In an AC setup if you wanted to be protected against a generator failure, you'd need two identical gensets, each large enough to run the whole load. With DC, say you had 5 generators but 4 could power the load. You still have no single point of failure, and you don't have to buy *double* the generating capacity.
Oh, and if a second generator fails, say you're down to 3, you're below the break-even point, but you're still limping along, with the operating generators assisting the batteries, extending your battery runtime long enough that you can probably fix one of the failed gensets. Oh, you found a spare generator at the rental place down the street? Switch a few rectifiers onto it and watch your charge status come back into the green. You just don't have that sort of versatility with AC.
DC is easier to noise-filter than AC. Keeping the high-frequency noise from switching converters off the AC input is something of a black art, and is hard to do effectively. You also have Power Factor (PF) issues when running large numbers of computers (or anything that uses switch-mode power supplies) from AC. Hence, your supplies have to be PF-corrected, which adds bulk and complexity, and reduces efficiency.
A DC-DC converter suffers none of those problems, going from your 48v battery bank down to the 12, 5, and 3.3 levels in your servers. It's easy to filter the switching noise because the input is DC, a big L-C filter works quite well. There's no such thing as power factor on DC, so the converters themselves are simpler and smaller, and run cooler.
One other huge benefit is that 48 volts is "low voltage" according to the NEC, so you can wire it yourself. You'll never have to let pole-climbers into your server room again.
Another advantage is that most DC-input equipment has a telco heritage, and supports dual inputs. Everything in telco has an "a-side" and a "b-side" power supply. It's only relatively recently that high-end datacomm gear has started to support multiple AC power inputs. History and experience are on your side with DC.
The DC step in the middle is so you can chop it into a high frequency square wave. You save power and space because you can vary the duty cycle of the square wave for regulation, and the high frequency allows for smaller components.
I've had enough abrasive sigs. Kittens are cute and fuzzy.
The first step will be in the home office. Have you taken a look at the rat's nest under most desks? Most of it is AC/DC conversion. If the industry could just arrive at a DC power standard, we could start with a single AC/DC "box" under the desk with a standard plug end for all DC peripheraps. Add daisy chaining and wireless USB or Bluetooth, and that nest is largely eliminated.
At this point, we could start to build it into houses and other buildings.
More
1) Contacts tends to rust on the positive side.
2) Lower voltage means bigger current for the same power. This would require thicker, more expensive cables
3) DC-DC voltage conversion is, somewhat less efficient... Ok, I know switching mode power supplies are efficient, but this leads me to the last point:
4) No insulation between systems. That way, systems get more prone to ground loops...
"There is always an easy solution to every human problem -- neat, plausible, and wrong."
H. L. Mencken
There are definite advantages to DC power... but it can also be *hugely* annoying.
I've worked in DC powered labs.
There isn't really any concept of 'plug' in the DC powered world. Powering up a device usually entails reading it's current draw off the equipment, selecting the correct gauge of wire, cutting the correct length of wire, strip both ends, hook up to your DC distribution on one end and your equipment on the other, select about the right size fuse, plug it in... etc. It's a royal pain. Oh, and make sure you do it correctly, because it's not that hard to electrecute yourself...
Nearly every engineer I've ever worked with whose been exposed to DC powered labs has begged to return to the AC powered world... it's just MUCH easier to work with.
On the flip side though... telco racks rock! Nothing beats hex head rack screws... you can literally drive them in at a 45 degree angle with a power drill and it's OK. It makes going back to the world of crappy philips head data wrack screws that you occasionally have to drill out because the head has stripped very annoying.
Wouldn't you still need a power supply to convert to the various voltages required by a computer, which may change over time. Most DC power setups I've seen run at 48v, which still requires conversion to 12v, 5v, and 3.3v. You can buy a 48v power supply for most servers and other equipment today. With a switched power supply, you'd need larger capacitors or a higher switching frequency in order to smooth out the lower powered DC. It's very unlikely that you would eliminate any heat loss. I would assume that telecommunications equipment uses a 48v setup due to legacy issues and that it was a better idea before switching power supplies became cheap and efficient.
.8 or so, meaning 20% of the billable power is effectively wasted and could be recovered for a slight increase in cost.
The downside with DC is that lower voltages require much thicker wires, and you're at much greater risk for fire. Circuit breakers and other things are also more complicated and expensive since DC tends to weld things together.
An EE I know just built a data center supplying 208v (2 branches of a 3 phase iirc) to all the racks. Almost all existing power supplies can take it, and it saves a bundle in wiring costs. I'm not sure about servers, but most desktop power supplies operate at a power factor of
The slashdot story intro implies that the advantage of DC is that you
save a conversion step. Well, maybe you do, maybe you don't, but
counting the number of AC-to-DC and DC-to-AC conversions is very
misleading.
Converting 50 or 60 Hertz to DC is much more costly and less efficient
than converting in either direction at a higher frequency. Low
frequency rectification requires large filter capacitors, complex and
expensive inrush current limiting, and active power-factor correction.
By doing that front-end work in one place only, preferably from a
3-phase source, you save power and increase reliability. You probably
still want multiple 50/60Hz to DC rectifier stages, of course, but now
they can be in parallel (for redundancy), rather than each one
downstream of the other where a failure of either one will bring down
the system.
Just because you're distributing DC to the racks, doesn't mean you
don't have to convert it again. It typically gets converted to AC and
back to DC at least once, usually twice before it reaches CPU and
memory chips. That's equally true in data centers that distribute AC
or DC. The fact is, memory and CPU devices want very low DC voltages
and very high currents. To make matters worse, not all parts of the
system want exactly the same DC voltage, you almost always have to
have multiple supply rails. You can't distribute very low voltages,
because it would require wires as thick as your arm and they'd still
be too resistive and inductive, so instead you distribute the DC at,
typically, 48 volts. The subsequent conversion to low DC voltages has
to happen via an intermediate AC, but it's a high frequency AC, so it
can be done much more efficiently using ferrite magnetic components,
active rectification, and often resonant mode filters. This high
frequency AC is confined to the internals of a power supply unit, it
never travels over wires or between boxes, thus reducing typical
high-frequency problems such as RFI.
I haven't mentioned battery-backup (i.e. UPSs). They make the system
more complex, but don't change any of the fundamental concerns. Even
on a DC distribution system, the UPS system requires it's own
additional stages of DC->AC->DC conversion, both while charging
(standby) and while discharging (during AC power failure). This is
because battery charging has to have a precisely controlled current
envelope. And batteries don't discharge at the uniform and
well-regulatted voltage that your DC distribution wants. They need
regulators, and switchmode regulators (typically DC->AC->DC) are the
most efficient choice.
Does big iron still use 3-phase power?
Yes. Mainframes, large UNIX systems, and the storage boxes that connect to them still require three-phase. (I am a storage specialist.)
SirWired
The article on raised flooring was an interesting question, but stupid solutions.
That article talked just like some "Intelligent-Design" moron. Just because HE can't figure out how to properly model raised-floor airflow, it must not be possible to do it at all. Wrong. There are any number of companies that will do this for you.
The solution to raised floor airflow is proper modeling of the equipment, vent tiles, and blowers, and relatively unobstructed floor plenum. The solution is NOT air-cooled equipment on bare floor and overhead cable runs. If cooling is still a problem, then use liquid-cooled racks and equipment. (This is where things seem to be going right now.) While overhead cable runs may work fine for some dinky test lab, "real" equipment requires power cables of a size that would quickly fill most overhead runs.
This article proposing DC power is equally stupid.
An enterprise storage box, fully configured that I looked at requires 13,800 kVA of 208V three-phase power (100A inrush current). My mind can barely fathom the completely unbendable copper "wire" that supplying that much juice at 40-ish volts would require.
Telco's switches have a far lower power density than modern servers, and the DC power was made to correct for different problems.
If this guy's ideal data center is overhead cable runs, ceiling blowers, bare floor, and DC power, I'd run away fast.
SirWired
Easy.
First, DC actually is better for transmitting power over long distances. AC current tends to concentrate in the surface of the conductor, leading to higher current densities and larger ohmic losses.
So, why do we use AC almost everywhere? Transformers. It is relatively easy and efficient to use a transformer to change voltages of AC power. For large electrical lines, the voltage is cranked way up, which means the current is reduced. The less current, the smaller the losses due to resistance in the wires. So power is transmitted at high voltages, so the current and hence losses are low. Then, near the place where power is needed, transformers change the power to lower voltage, higher current. (This is because you can't have house wiring and appliances that won't arc or explode when hit with 13,800 V.)
Converting between high and low voltages with DC power is much more difficult, and requires more complex equipment. (An AC transformer is two pieces of wire wrapped around a chunk of iron.)
I think this article might be using the term "engineer" a bit too loosely. I doubt any company would hire an engineer - an actual person with a Professional Engineer's License - to work on these systems. A more appropriate term might be "technician," which usually refers to someone who is trained to repair and work with a single type of technology. Engineers, on the other hand, are usually trained to work with a large variety of technologies and usually work on either (A) Research and development, (B) Manufacturing, or (C) Failure analysis and redesign.
I guess using the term "engineer" sounds better though since it tends to scare the corporate fat-cats away from a technology because of the implied additional cost from hiring an engineer as compared to a technician.
Yes, real big iron still uses 3-phase power. I can only speak on behalf of large IBM system (zSeries, etc). These systems will accept 192VAC to 508VAC on the input, 3-phase Delta. This means no neutral required. Additionally, they will even run with one phase totally missing. The first power conversion stage in any piece of their 'big iron' is a very large AC to DC converter, rated for a 350VDC output at over 42kW. Actually it's six 7.5kW converters paralled, and these are redundant/hot swappable. Totally modular, with no cable connections. This block is about 95% efficient. This DC is then distributed to the rest of the system power supplies, with redundant cabling supplying all point of load converters. All point of load converters are also redundant and hot swap. These converters have a range of efficiencies, but are typically much better than industry standards. A DC/DC converter in the z9 can source 1000A alone on the CPU Vcore level (12 of these supplies are in the machine). Supplies are used for CPU nodes, I/O cages, blowers and refrigeration. All blowers are 3-phase DC-brushless type, with the 3-phase synthesized off the 350VDC feeds. The blowers are usually 300W or larger, each. The CPU refrigeration is also run by 3-phase compressors, this power also being synthesized off of 350VDC. This is done to allow a conventional off-the-shelf compressor to be run off any line voltage, and ride through phase losses (as this is seen by the bulk AC/DC converter instead). The 'big iron' also supports built in UPS cabability, allowing you to connect battery packs directly to the bulk AC/DC converters. A machine will handle six 400V@2.5Ah battery packs connected to it. This feature is used to ensure a system such as a z9 has true 100% availability, and won't suffer a hard shutdown due to careless datacenter workers or electricians. In short, the article is intend to address small white box systems that use $12 power supplies with very poor reliability and efficiencies. And to another poster that brought up 3-phase being more efficient for power conversion...that's not really true these days, as everything requires power-factor correction. Nothing in the IT uses huge three-phase bridge rectifiers and phase-regulated primaries anymore.
Yes, real big iron still uses 3-phase power. I can only speak on behalf of large IBM system (zSeries, etc). These systems will accept 192VAC to 508VAC on the input, 3-phase Delta. This means no neutral required. Additionally, they will even run with one phase totally missing.
The first power conversion stage in any piece of their 'big iron' is a very large AC to DC converter, rated for a 350VDC output at over 42kW. Actually it's six 7.5kW converters paralled, and these are redundant/hot swappable. Totally modular, with no cable connections. This block is about 95% efficient.
This DC is then distributed to the rest of the system power supplies, with redundant cabling supplying all point of load converters. All point of load converters are also redundant and hot swap. These converters have a range of efficiencies, but are typically much better than industry standards. A DC/DC converter in the z9 can source 1000A alone on the CPU Vcore level (12 of these supplies are in the machine). Supplies are used for CPU nodes, I/O cages, blowers and refrigeration.
All blowers are 3-phase DC-brushless type, with the 3-phase synthesized off the 350VDC feeds. The blowers are usually 300W or larger, each.
The CPU refrigeration is also run by 3-phase compressors, this power also being synthesized off of 350VDC. This is done to allow a conventional off-the-shelf compressor to be run off any line voltage, and ride through phase losses (as this is seen by the bulk AC/DC converter instead).
The 'big iron' also supports built in UPS cabability, allowing you to connect battery packs directly to the bulk AC/DC converters. A machine will handle six 400V@2.5Ah battery packs connected to it. This feature is used to ensure a system such as a z9 has true 100% availability, and won't suffer a hard shutdown due to careless datacenter workers or electricians.
In short, the article is intend to address small white box systems that use $12 power supplies with very poor reliability and efficiencies.
And to another poster that brought up 3-phase being more efficient for power conversion...that's not really true these days, as everything requires power-factor correction. Nothing in the IT uses huge three-phase bridge rectifiers and phase-regulated primaries anymore.
We seem to have moronic moderators today, as it's uncomprehensive why the above comment got modded as "Insightful".
For Christ's sake, this guy doesn't know what he's talking about! "DC power is naturally unstable", "unclean power" WTH?
Back to the original topic, the article is, as other mentioned already, 100% pure dribble. The major advantage of AC input power is that the power conversion (AC to DC system and from there down to 5V/3.3V/VCore/DDR/IO/etc), happens close to the loads.
AC voltage is 110V or more, therefore a power of 400W per system will give about 6.3A RMS (considering a conversion efficiency of 80%). So, one must design the wiring to the system to withstand let's say 7.5A. And then the last down conversion stages start from 12V only.
If you want to carry a DC voltage (e.g. -48V), then you must use an isolated DC/DC down to the system voltage because it's impractical to downconvert from -48V to every rail voltage needed in the system. The same 400W will be 9.3A out of the -48V input (for a 90% conversion efficiency).
Transmission power losses increase with the square of the passed current. The 9.3A versus 6.3A means one must use thicker wire, i.e. the wiring is costlier to carry the DC voltage and there is no obvious benefit.
In the end, if anyone wants to lower the electricity cost, he/she must invest in better off-line power supplies. For high power ones with PFC, cheapies have 65-70% efficiency, while a good design has more like 85-90%. This difference is significant.
Serban
Hi everyone,
I am a datcenter manager that has had the opportunity to not only run but also build a datacenter from pretty much scratch. In my experience I have found that both DC and AC powered equipment both have their places in the environment. Neither system is perfect so by running hybrid you can get the most flexibility.
We recently moved our datacenter form a 10K sq ft facility down to a 1700 ft facility by doing a technology refresh and changing many of our key infrastructure methods. In the new facility I currently have 315 HP blade servers plus another 10-15 traditional rack type servers running. I have the capacity to add up to another 144 blades (assuming they are 1U) before I run out of floor and HVAC capacity. The power delivery method is hybrid. I run DC for the blades which are fed by Emerson Energy's Candeo XL rectifier stacks (originally designed for telco) and AC for everything else. To eliminate a lot of the under floor clutter I use a trough system instead of conduit for the various AC circuits. HVAC is provided by 4 Liebert 22TON units which keep my room at a comfy under floor temp of 66 degrees.
Adequate airflow is critical so we spent a lot of time planning tile placement. The key for proper cooling in this scenario was a high volume of airflow pushing the cooling to about 5.5ft up from the raised floor. This way my cooling isn't being sucked up by just the bottom half of the rack. Low voltage cabling is overhead.
We chose to power the blades DC for two reasons. First was the limited space I had for installing breaker boxes on the walls. The number of AC circuits I had was limited so I pulled fat feeds directly to the Candeo systems. A full rack of HP p-class blades would require 4 x 3phase 208 circuits per rack. My initial installation of blades would have consumed 144 of my 168 circuits leaving next to nothing to power my SAN/Network/Tape Library/etc equipment. The other reason was power supply efficiency. In the conversion of power from AC to DC the efficiency of the power supply must be taken into consideration. It's not just the number of conversions you do but the loss at each. Typical power supplies in servers run about 80% efficient while my Candeo as they are setup gets about 90%. For me this ultimately meant less heat and more available cooling, therefore I could bring in more servers under the existing HVAC.
I prefer a best of class mentality. IMHO there is no best universal solution. For those of you that use traditional rack mounts servers like Dell you can purchase these units with a DC option. I am not sure if HP offers a similar option but they might.
Len
p.s. Someone also made the comment about DC not generating noise in network cabling while AC does. This is not a totally true statement. Anytime you run a current through a conductor you will generate a magnetic field. Put this in parallel to another conductor and you will further induce a mag field (this is why any power runs that have to intersect low-voltage cabling should only intersect at 90 degree angles to avoid inductance). The big difference is the way DC cabling runs. In most DC circuits the feed and return lines run together so the proximity of the out of phase magnetic fields will cancel each other out. Don't believe me? I had this problem when we intially wired these Candeo systems up. The small feeds to the racks and the big mains that connected to the common buss bar were about a foot apart. Because the fields weren't cancelling, we were getting enough noise on the lines that it looked like there was AC leaking through the circuits (6volts p-p in some cases). By simply wire tieing the lines together, the proximity cancelled the fields out and everything was peachy.