Power Consumption and the Future of Computing

mrdirkdiggler writes "ArsTechnica's Hannibal takes a look at how the power concerns that currently plague datacenters are shaping next-generation computing technologies at the levels of the microchip, the board-level interconnect, and the datacenter. In a nutshell, engineers are now willing to take on a lot more hardware overhead in their designs (thermal sensors, transistors that put components into sleep states, buffers and filters at the ends of links, etc.) in order to get maximum power efficiency. The article, which has lots of nice graphics to illustrate the main points, mostly focuses on the specific technologies that Intel has in the pipeline to address these issues."

Comments

For Funzies

[Mateo_LeFou]

anyone ever compared power usage (say, per flop) advances to the power usage (per mile) advances of our innovative combustion engine indutry?

Re:For Funzies

[UbuntuDupe]

anyone ever compared system failure (say, per computer-hour) improvements to the system failure (per mile) improvements of our innovative combustion engine industry?

anyone ever compared computer system transparency (say, how much of it has publicly available documentation) to the system transparency of our innovative combustion engine industry?

Re:For Funzies

[Anonymous Coward]

You'd probably find that computers are both more transparent, and have a lower failure rate.

That is if you compare per hour, per mile would be out by a huge factor.

Re:For Funzies

[dbIII]

Yes - it comes to a yearly rate of 37 Volkswagens per football feild.

Re:For Funzies

[Anonymous Coward]

If only it weren't for those engineering hurdles, we'd be running our engines on water.

Re:For Funzies

[Nullav]

That's like comparing the number plane collisions per year to the number of car collisions per year. (For one, a CPU doesn't move in any meaningful sense.)

Re:For Funzies

[AngelofDeath-02]

I looked up water engines. They had a model that used salt water and some cheap comodity metal to produce hydrogen.

My guess is that metal is lead, no?

I would find it ironic that we'd look at running a combustion engine off of a battery in such a direct form ;)

Multipurpose.

[Anonymous Coward]

"I said in my recent Griffin/Puma article that the key to power optimization lies in turning off components when they're not in use."

I should point out that they're are chips that dynamically reconfigure themselves to do different jobs as required.

Re:Multipurpose.

[grimdawg]

I should point out that there are words that sound the same, but dynamically reconfigure their spellings to do different jobs as required.

1000 watt power supplies

[bl8n8r]

With the availability of PC power supplies* in excess of 1000 watts, and the mine's-bigger-than-your's demographic, I wonder what bearing this has on power consumption also. Perhaps peripheral manufacturers need to concentrate on power usage also.

[*] - http://www.newegg.com/Product/ProductList.aspx?Sub mit=ENE&N=2010320058+1131428171

Re:1000 watt power supplies

[MP3Chuck]

If you look at the number of reviews, though, compared to something like a more modest 500W PS it would seem that not too many people really use/need a 1KW PS.

In fact, those high-end 1KW supplies might even be better for power consumption since they tend to have higher efficiencies than the cheapo options.

Re:1000 watt power supplies

[Anonymous Coward]

Yes, a several hundred dollar 80-85% efficient 1kw power supply will probably use a lot less power than a crappy $15 600w one. On the other hand, the octocore overclocked CPU, 4 hard drives and dual video cards will more than make up the difference.

Re:1000 watt power supplies

[ncc74656]

With the availability of PC power supplies* in excess of 1000 watts, and the mine's-bigger-than-your's demographic,

WTF actually needs that kind of power? I've built 16-disk 3U RAID arrays that don't use nearly that much power. Each is powered by a 650W RPS (made up of three hot-swappable 350W power supplies, capable of running on two if one fails), and actual maximum power consumption (measured with a clamp-on ammeter and a power cord with one of the wires pulled out in a loop) was somewhere around 350-400W while the drives were spinning up. Idle power was a fair bit less.

Let's virtualize!

[gelfling]

The problem is the massive rollout of servers and blades into datacenters that under utilized. The One Job One Box Syndrome. Most DC hardware is barely used but there it sits idling and sucking power. If we leaped into massive virtualization we'd be able to reduce the number of physical components and save power.

Re:Let's virtualize!

[holysin]

Yup, only downside would be the single point of failure for say, all of the company's servers. One job/one box has its benefits. Optimizing power and hardware utilization is not one of these benefits. (Of course if you do decide to utilize virtualization, odds are you're smart enough that if your business requires 24/7 uptime, you have a hot swap server serving as a primary backup which would be easy for you to roll out...) Power consumption is not on the radar for most companies now (excluding the huge server farms, they've noticed that power does cost money. So as of yet there's no push in corporate America. If the US goes through an (electricity) Energy crisis then things would change, but unless electricity in the States starts costing significantly more, I don't see virtualization taking off for power reasons...

Re:Let's virtualize!

[Repossessed]

Virtualization can be used to increase redundancy actually. You can set up virtualization across a number of machines, then set it up so that if one (physical) machine fails, the VM simply shifts the load to a different physical machine (probably a bit intensive on hard drives because of the additional redundancy, but then, it's hard to have to much redundancy for your data anyway in a business environment). The big catch to virtualization is gonna be the TCO if you want to use a proprietary OS. Either you get charged for each copy you run, or you get charged based on the number of processors you have. (2300 per socket for datacenter, quite probably more than the sockets themselves). Microsoft isn't even the bad one here, a couple of companies have been setting up VM price schemes in such a way that they get every last dime you should have saved by needing less physical machines. Microsoft is trying to make it at least a little cost effective though.

Re:Let's virtualize!

[WuphonsReach]

Yup, only downside would be the single point of failure for say, all of the company's servers. One job/one box has its benefits.

Typically, if you're going to virtualize - the minimum number for physical boxes is probably 3. During normal operations, you run your load spread across all 3 boxes, with the option to consolidate down to 2 boxes if one goes down. You can do it with just 2 boxes, but it's not going to be as nice. Naturally, if you have the server load to require 4+ boxes, it becomes much easier to move things around.

Virtualization is definitely not a cure-all. It requires some forward thinking and planning to make sure that you can handle (or avoid) outages by having redundancy built-in to all aspects of the system. Everything from power, UPSs, head servers, the switch fabric, network cabling, backups and SAN units.

(Currently moving towards a 3-4 box setup for our company. It'll cost us around $65,000 over 3 years to complete the migration to a virtualized environment.)

Huh. Well I never.

[Colin Smith]

Computer performance being limited by power. Who'd have thought.

Big cuts

[Forge]

The thing with power usage is that nobody seems interested in attacking the 2 largest areas of power wastage. (except maybe google)

#1. DCAC conversion.
Your typical Datacenter has a UPS or batteries and inverters (Enterprise scale UPS). What this amounts it is AC power from your utility company converted to DC for storage in a battery and then converted back to AC to supply the Server's power supply, then converted back to DC to actually run the components of the computer.

Ever notice how hot a UPS gets during normal operation? That's power going to waste. The solution is to run our servers at a standardised DC voltage. 48 Volts sounds good since that is already defined for Telecom equipment (correct me if I'm wrong. I am not sure of the figure)

#2. Raised flour and underground AC. A good chunk of datacenter power is used to run the air conditioning. If we abandoned the notion of raised flours and replaced them with say insulated celling mounted ducts with vents faceing each rack.

While we are at it here is another simple power tip. Turn your rows of racks back to back. When they all face the same direction, hot air blows from the back of one machine to the frunt of another, forcing the AC to work overtime. In my design, I would have extraction fans betwean my back to back racks, pumping the hot air outside (or into the office during winter. For those of you who have winter.

Re:Big cuts

[johnw]

#2. Raised flour and underground AC. A good chunk of datacenter power is used to run the air conditioning. If we abandoned the notion of raised flours and replaced them with say insulated celling mounted ducts with vents faceing each rack. ITYM self-raising flour. Raised flour is a cake.

Re:Big cuts

[NeverVotedBush]

Lots of errors in your suppositions.

DC/AC conversion? The bigger data centers can't use batteries - too many, too big of a hazard, etc. They use rotational UPS's. These stay AC all the way.

Additionally - power distribution is better at higher voltages. It's that current squared thing. More and more equipment is also going to higher voltage distribution on the boards with local DC/DC conversion at the load. For the exact same reason. Our center distributes at 208 volts.

The argument against a raised floor is bogus. That acts (and is necessary) not only for cabling, but also for air distribution. Heated air rises. Feeding cold air up from the floor to where it flows into the racks to be heated and then recovered at the ceiling is the most efficient way for air. The fact that the floor is not insulated is a non-issue. The whole room is being cooled. The temperature is the same on either side of the floor tiles.

And about the face to face and back to back layout of racks - every single one of our racks is already in that orientation for exactly that reason. We have hot aisles and cold aisles and the temperature difference between them is pretty marked.

The next wave is a move back to "water" cooling. Either plumbing liquid to each rack where in the rack it locally grabs heat from circulated air within the rack, or plumbing into the boxes themselves. This is simply because heat loads are going up and it gets harder (and louder) to pump enough air through a building to cool the more dense newer equipment. Plus people don't have to put on jackets to go out on the floor or yell to be heard in a big data center.

Re:Big cuts

[Firethorn]

I've seen raised floor AC done right. Each rack was sealed, had a vent in the bottom and a vent into the ceiling. The AC pushed cold air into the subfloor, which was then sucked into the racks, with hot air rising into the ceiling. Where the AC pulled the hot air to be cooled again.

Also, 99% of UPS units don't convert AC to DC unless it's charging the batteries. Normally this would only be a trickle charge. If the UPS is providing power, you're in a critical situation anyways, I wouldn't worry about the fact that a UPS isn't particularly efficient, as you're probably spending 99% of your time not on UPS.

As for switching to telephone industry standard 48V power, you'd be converting it again to whatever the equipment wants, much of it 12V or less. 120VAC->12VDC is more efficient than 120VAC->48VDC->12VDC. In addition you run into the problem that 120VAC over 12gauge cable wastes less than half of the power that the same wattage of 48VDC would waste over the same diameter cable. So you'd have to use heavier gauge cable - payback isn't quick for that by any means.

You might be able to get away with it on a rack level, powering all the blades on 48V via rails to a couple of redundant power supplies somewhere in the rack. Either top or bottom, depending upon cooling and other requirements, though the middle might be an interesting choice, as it'd allow you to have half the wattage running over the rails on average(you'd have two runs instead of one).

You want to save power? I'd switch to feeding the racks/power supplies with 240V lines. Half the line resistance for the wattage.

Re:Big cuts

[Attila Dimedici]

"While we are at it here is another simple power tip. Turn your rows of racks back to back. When they all face the same direction, hot air blows from the back of one machine to the frunt of another, forcing the AC to work overtime. In my design, I would have extraction fans betwean my back to back racks, pumping the hot air outside (or into the office during winter." Pumping the hot air outside might help depending on outside temperatures. But, if the air you are pumping outside is 80 degrees F and it is 90 degrees outside, you would make your AC work harder (the air you are replacing the 80 degree air with is 90 degrees). Placing the units back to back rather than front to back will have negligible impact (it will only make a difference if the hotter air entering the second unit causes the second unit to use more power, which would then in turn generate more heat, causing the AC to use more power).

Re:Big cuts

[NeverVotedBush]

It actually does not force the AC to work overtime. You are merely blowing hot air into the next rack to cool it - which it does less effectively. The parts in the second rack just run hotter. Overall, the AC actually will ultimately carry the same load. Your boxes will just overheat. The heat load on the room is just the power you supply to your boxes. They are, in effect, nothing but space heaters.

Data centers recirculate the same air generally because it is cheaper in the building design, more reliable, etc. Plus the AC has to control the humidity as well - not easy to do if you are consatantly changing your source air supply, its temperature, and its humidity.

As to rack cooling, you also have to watch out for eddy flows around the ends of rack rows and over the tops as well. If the room is not well designed, end racks in a row will run hot due to air short circuiting and just rolling around from bact to front again. Over the top is bad too and a reason high ceilings help.

Re:Big cuts

[pla]

The thing with power usage is that nobody seems interested in attacking the 2 largest areas of power wastage. (except maybe google)

Except, those don't really waste most of the power.

With AC/DC, you already have equipment available that can push over 90% efficient. With air conditioning, central home units manage 90-94% efficient, and I'd expect industrial models to do even better. So not a lot of room for improvement there.

With servers, however... The better they scale to their load, the more efficient they get. In theory, depending on your required maximum transaction time, you could actually suspend the whole machine giving very nearly zero power consumption during offpeak loads. In practice you can't really do that (completely waking up can take half a minute even under ideal conditions), but using ondemand CPU and memory speed scaling, halting all but one core, and spinning drives down (but not stopped), you can still get huge power savings (just for the CPUs in an 8-way Xeon machine, that drops you to literally 2-5% of the power draw!).

Of course, improving all three gives ever better savings, but it makes more sense to focus on the easy targets first. And in this case, the servers themselves present the easiest targets.



While we are at it here is another simple power tip. Turn your rows of racks back to back.

Or better yet, just line them all up in a single long row, such that a wall divides the front from the back, with the back exhausting outside (or into the offices in the winter, as you suggest). That way you can still get to the back when needed, just by going into the other half of the room. I have to agree, unless you have a datacenter in the desert, I don't understand why we AC them instead of just pumping as much outside air as possible through the room. Moving a large volume of air (given a low pressure change) costs next to nothing compared to actual air conditioning.

Re:Big cuts

[DaleGlass]

Also, 99% of UPS units don't convert AC to DC unless it's charging the batteries. Normally this would only be a trickle charge. If the UPS is providing power, you're in a critical situation anyways, I wouldn't worry about the fact that a UPS isn't particularly efficient, as you're probably spending 99% of your time not on UPS.

That's a cheap, consumer oriented UPS. Datacenters use the kind described, ones that are always doing the AC -> DC -> AC conversion. What this achieves is that instead of the UPS taking over when the line voltage isn't good, the UPS is always providing clean power because everything goes through it. One of the advantages is that this kind has no delay for switching between AC and battery, as there's no switching involved.

Re:Big cuts

[Courageous]

Or better yet, just line them all up in a single long row, ...

*cough*

I think our data center might be bigger than your data center...

C//

Re:Big cuts

[stonecypher]

Also, 99% of UPS units don't convert AC to DC unless it's charging the batteries. Normally this would only be a trickle charge.

That trickle adds up. Try doing the math for a datacenter with 100,000 blades (relatively small,) which would suggest at least 20,000 UPSes. Consider that that trickle comes in at about 1.5% inefficiency, and look at the end result. You might be surprised.

Re:Big cuts

[jbengt]

? 20,000 UPSs ?!?
Try 2 to 4 big ones.

Re:Big cuts

[jbengt]

"you're probably spending 99% of your time not on UPS."

Almost by definition, you're always going thru the UPS; what you're not doing 99% of the time is discharging the batteries.
And a large, efficient UPS is proably only around 90% efficient at normal loads.
At very low loads, they can actually use more energy than at full loads.
So a 250 kVA UPS is going to turn about 20 to 25 kW of energy into heat, even when the equipment it's serving is idling.

Re:Big cuts

[mblase]

Ever notice how hot a UPS gets during normal operation? That's power going to waste.

Maybe for you. I've been using mine as a nacho-cheese warmer for months.

Re:Big cuts

[jbengt]

"With AC/DC, you already have equipment available that can push over 90% efficient"
Yes, but when UPSs are designed for maximum load, and redundant UPSs are installed, and you typically are operating below 50% of capacity (e.g. late shifts), that 90% full load efficiency can be below 50% real life efficiency.

"With air conditioning, central home units manage 90-94% efficient, and I'd expect industrial models to do even better"
Not even close, if you assume that you're talking about 90-94% of theoretical maximum cooling effect. A common EER (btuh's removed divided by watts consumed . . . yea, I know, US units are crazy compared to metric) is around 10 to 13. That's a Coefficient of Performance of around 3. So you're removing 3 times as much heat as you're putting in, but the theoretical limit would be a COP of around 6 or 7 at typical temperatures.

Re:Big cuts

[WeblionX]

I have to agree, unless you have a datacenter in the desert, I don't understand why we AC them instead of just pumping as much outside air as possible through the room. Well, there's the whole filtering the air and removing humidity and dust thing...

Re:Big cuts

[scottv67]

I don't understand why we AC them instead of just pumping as much outside air as possible through the room.

What do you do on the days when the temp outside the building is below 32F? How about when the temp does not go above zero degrees F for a couple of a weeks in the middle of winter? Do you know what happens to the relative humidity of air that is heated from zero degrees F to 70 degrees F? Are you going to spend a lot of money for equipment and power to humidify that air as you pull it into the data center and then shoot that nice humid air back outside?

Re:Big cuts

[anti-human 1]

I thought he was on to something, running your power conduit in flour to insulate heat from the rest of the datacenter. :P

Unfortunately for me, I got back from Costco -AFTER- reading your reply explaining its a typo. I was going to reinsulate my attic. Anyone have a use for two pallets of flour?

Re:Big cuts

[Skapare]

Unfortunately, in the USA, power to most commercial/industrial buildings is not available in 240 volts. Power on a large scale is provided in three phase so the power company distribution is kept in balance. But in the USA the primary choices for that are 208/120 volts, or 480/277 volts. Many power supplies would probably work OK at 277 volts, but since they are not specified for that, it's risky from many perspectives. 208 volts would work for full range power supplies, but maybe not for those that have two switchable voltages. Some parts of the USA can get power at 240/139 or 416/240 volts. That would work for all those international power supplies that can do everything from 100-240 volts.

Re:Big cuts

[Firethorn]

As the other poster noted, you'd only need 20k UPS units with 100k blades if you're using small UPS units, not 'building' level ones that sit somewhere else and have whole racks of batteries.

As for the 1.5% efficiency - The larger the UPS, the more efficient the charging system. Still, you can't get away from the fact that you need a float charge for lead-acid batteries, indeed, for most rechargable technologies. Still, the number of batteries needed depend on how many kwh you need to store. For 100k blades, your UPS system isn't going to be meant to last long, only enough time to get the generators spun up and taking the load. Maybe a half hour. You're going to have multiple generators as well, in case one of those fail.

Re:Big cuts

[Firethorn]

Unfortunately, in the USA, power to most commercial/industrial buildings is not available in 240 volts.

Huh, I've seen it available in most buildings I've been in. Even so, as long as it's AC, you can efficiently transform voltages around, even if you need a big transformer in a mechanical room somewhere.

120Volts doesn't come in on it's own set of wires, it's set up as a split phase via grounding from two 240 volt lines.

My general point is that it's more efficient to move high voltage around than low voltage.

Re:Big cuts

[Forge]

Any computer I can run for extended periods in a none air conditioned building here (Jamaica) can be considered a solid desktop. Haven't met server like that yet. An AC breakdown at the datacenter is a major crises. The repair crew on that system has 30 to 90 minutes to repair the C before servers start dropping.

In other words. Add "tropics" to "desert"

Re:Big cuts

[Forge]

Hmm.. Reminds me of that Pentium egg oven :)

Re:Big cuts

[redcane]

Warmer air can carry more water right? So the air inside the data centre would be extremely dry. Most computer equipment specifies 0-x% humidity. So lowering it doesn't adversely effect the equipment. I'm sure the outside environment doesn't care either.

Re:Big cuts

[vuffi_raa]

on #2- I agree with you, but fire marshalls don't see it that way. We have 3 datacenters in my company and airflow and room temp are strictly monitored by the fire marshalls. If they look at the room and say it's not how they want it they shut down the entire company until compliance is made.

Re:Big cuts

[evilviper]

AC power from your utility company converted to DC for storage in a battery and then converted back to AC to supply the Server's power supply, then converted back to DC to actually run the components of the computer.

There are many groups that have expressed interest in DC datacenters.

The reality, however, is that AC/DC conversion is only nominally less efficient than DC/DC conversion. With the increasing popularity of 80plus efficient PSUs, there's very, very little to be gained by going to DC. You're really talking about significant investment to get a ~5% improvement in efficiency.

Ever notice how hot a UPS gets during normal operation? That's power going to waste.

All devices, everywhere, get hot and waste energy. Some waste more than others, of course, but feeling how hot something gets won't tell you that. How hot something gets just tells you how under-engineered their respective cooling systems are...

For example, I've got a 4W amplifier that will burn your hand, a 20W soldering iron that will vaporize skin, and 60W TV that barely feels warm to the touch.

If we abandoned the notion of raised flours and replaced them with say insulated celling mounted ducts with vents faceing each rack.

I would go the exact opposite direction... Keep the room cool, and use ducts to remove the hot air (eg. to the outside).

I would have extraction fans betwean my back to back racks, pumping the hot air outside (or into the office during winter. For those of you who have winter.

See above. Unless you carefully duct the hot air from the source, you run a big risk of removing just as much cool air as warm, and the warm air will still seep out and get away from you, and raise the temperature of the room.

I don't run a data center, but I have tried this, first hand, on a smaller scale.

Re:Big cuts

[scottv67]

Most computer equipment specifies 0-x% humidity. So lowering it doesn't adversely effect the equipment.

Ok, you go ahead and run your data center with 10% relative humidity. Don't be surprised when static electricity becomes a big issue for you. Also, I doubt that you find that "most computer equipment specifies 0-x% humidity" when the equipment is running (you might be able to *store* the equipment at 10 percent relative humidity). There is a reason that most data centers are kept at 40-45% relative humidity year-round.

Also, very low relative humidity is hard on the respiratory systems of humans working in that environment. I speak from first-hand knowledge. Even if you could keep your machines running at 10% relative humidity, people would not want to spend long periods of time in your mega-sized food dehydrator.

Re:Big cuts

[Forge]

It took a while but a logical fact infested response to each point I make.

I won't go point by point because much of what you say amounts to a command for me to do more in-depth research before claiming either of us is right. And some of it makes me go "gee, I didn't see it that way". (The ducts for hot air spring to mind)

However to the wattage. The Solder iron isn't wasting energy. Generating heat is what it dose. The 60 Watt TV is cool BECAUSE it is efficient. Most of It's power is being used to put a bright image on your screen.

Now to DC/AC. In my world Each rack would have a small UPS without an inverter that feeds equipment in that rack alone. Inside the Servers, DC to DC conversion is going on anyway so dropping the losses from AC to DC conversion can't help but be a net gain.

Re:Big cuts

[evilviper]

The 60 Watt TV is cool BECAUSE it is efficient. Most of It's power is being used to put a bright image on your screen.

I think you just misunderstood my point. A 60watt TV may be cool to the touch, while a 30watt TV could be extremely hot to the touch.

There are two parts to this:

1) It's any use of energy, not just WASTE of energy that makes heat.
2) How cool a device stays has very, very little to do with how much energy it is using/wasting, unless they're identical in every other way (which basically never happens).

Inside the Servers, DC to DC conversion is going on anyway so dropping the losses from AC to DC conversion can't help but be a net gain.

That's not necessarily true. Certainly several of the numerous voltages output by a normal PSU are accomplished by DC-DC conversion, but it's entirely possible that the two major voltages (5v and 12v) could be converted directly from AC with, eg. a transformer.

And still, even if you have a net gain in efficiency, economies of scale are against you. 120/240v AC ATX PSUs are so standard, and have been around so long, that they are often given away for free, or very near it. Meanwhile, if you need any other kind of 300W power supply, you can expect it to cost around $100. It's actually a depressing state of affairs.

And I happen to know from pricing a few of them, that DC ATX power supplies are significantly more expensive than their AC brethren. So, at best, going all DC for energy efficiency would be a case of diminishing returns, and plenty of drawbacks as well.

Re:Big cuts

[Forge]

The relative cost of DC PSUs is (as you mention) a function of market forces. Specifically volume production. This however can be changed over time. Hate to draw a car analogy but I have to.

In most places, spare parts for different Japanese sedans are pretty close.

In Jamaica, that was the case until the Police Force standardised on the Toyota Corolla over a decade ago. These days Corolla spare parts cost a fraction of Honda spares.

Yeah. It's depressing. No Data centre customer has large enough needs and influence to resemble a small islands police force.

Re:Big cuts

[Skapare]

Most commercial/industrial buildings have 3-phase power. Most 3-phase power is of the "star/wye" configuration, which means 3 separate 120 volt transformer secondaries wired to a common grounded neutral. At 120 degrees phase angle, the voltage between any 2 of these 3 lines is 208 volts, not 240 volts. There are some exceptions where commercial buildings get single phase power, or an older delta type 3-phase system that has various kinds of problems with it.

They will use standard 240 volt outlets for that 208 volt power and some things will even work on it. Motors will draw more current and could overheat (if designed for the current they would have at 240 volts only). Heating elements will run at 75% capability (wattage). Computers designed for wide range (100-240 volts) will work. Computers designed for split range (110-120 and 220-240) may work at 208 volts.

Yes, the higher voltage is better because the current is less and the voltage drop/loss as a result of that will be less, even if the wires are thinner in proportion to the reduced current. It's just that you don't quite get as much of that as you could when you have 208 volts instead of genuine 240 volts.

Google Distinguished Engineer's point of view

[naeim]

A couple of months ago, Luiz André Barroso of Google gave a talk at Stanford about this very topic. Unfortunately the talk wasn't recorded, but here's a summary: http://cs343-spr0607.stanford.edu/index.php/Writeu ps:Luiz_Andr%C3%A9_Barroso

Laptops in the datacenter

[brunes69]

I have been wondering this for a while now.

Why can I sit here and type this on a laptop that is faster than a top-of-the-line 1U rack from 1 year ago, and yet data centers are still loaded with power-sucking 3 year old machines by the thousands?

What you need in a data center is a) Performance, and b) Reliability. Performance is already covered - every year laptop speeds match the top speeds of the previous year's desktop machines. So you're at most a year behind the times. As for reliability - anyone who works with data centers knows that reliability does not come from reliable hardwaye - it comes from redundancy. If you had a data centre built from laptops in a blade-style configuration, where any one could be swapped out at any time, you would have awesome reliability at a tiny fraction of the power useage.

Throw in some external iSCSI or ATAOE storage vaults, and you're good to go.

Re:Laptops in the datacenter

[ZorinLynx]

Or perhaps not laptops, but machine designed using laptop components.

I've heard of folks using Mac Minis as servers. They use laptop mainboards and hard drives, so they consume very little power, but are plenty fast for a lot of server needs.

-Z

Re:Laptops in the datacenter

[Firethorn]

What do you think that blades generally are? Well, not exactly, they use components that are designed to be used 24x365, faster memory subsystems, multi-processor, etc...

As for replacing machines every year - that's a big false economy. Even cheap power hungry servers cost more than their electrical costs per year*. Then you have all the hardware concerns and swapouts.

Swapping out all your servers in a farm annually would be a good way to get the greens coming down on you.

*Assuming sane prices per kwh, of course.

Re:Laptops in the datacenter

[NeverVotedBush]

"Why can I sit here and type this on a laptop that is faster than a top-of-the-line 1U rack from 1 year ago, and yet data centers are still loaded with power-sucking 3 year old machines by the thousands?" Uh... Cost? How much would thousands of your laptop cost to replace them every year to keep up with the latest technology?

Like most computer buyers, you get on that treadmill when you have to. At first you have the fastest machine on the block (depending on your price range), but then as it gets older, newer equipment is faster and costs less. But only fools spend money like crazy to stay on the crest of that curve.

Plus it costs a bundle to rip out thousands of computers, possibly rewire to meet new power desity requirements, and then bring in the new boxes, bring up the new system, run acceptance tests, reconfigure software so old code runs on the new boxes (changes in OS happen too you know), retrain for maintenance, and then finally put the new boxes into production.

There are huge costs in the machine, infrastructure, down time, etc. That's why. You need to have a system up long enough to get useful work out of it before you chase the next innovation.

Re:Laptops in the datacenter

[stonecypher]

As for reliability - anyone who works with data centers knows that reliability does not come from reliable hardwaye - it comes from redundancy.

It comes from both. Also, I'm not sure why you believe laptops perform in comparison with 1U units from the previous year; there's a lot more to speed than processor type, the mobile processors don't compete per megahertz with the standalone processors, and those laptops cost something like 3:1 for the same hardware statistics.

If what you said was true, these various datacenters would have switched long, long ago. Turns out the people running places like Level3 aren't idiots.

Re:Laptops in the datacenter

[prestorjohn]

plus, you could sell the laptops (thru middle-man, donate to schools, give to lowlevel employees, whatev) when thru a lot bigger market for old working laptops than old rackmounts

The future computing device uses less than 10W

[Anonymous Coward]

The future of desktop computing is 24/7 thin clients/home servers using less than 10W and passive cooling without fans, because for a typical 300W desktop 24/7 system you probably would be paying $100/month, more than a thousand a year. This is enough for 90% of users, those who are not after the latest/greatest 3D horsing power, those whose necessities are supplied with an onboard graphics chip such as Intel X3100 or even less. You would be surprised with the amount of computing power such devices have nowadays.

They do not use hard disks, but flash memory/pendrives as storage for the operating system and homedir, and are passively cooled, so they do not use fans, which are noisy and spend more energy. Massive storage (TBs) can be added if necessary, each one using an extra ~15W. A small list with some of them:

1) Linutop: http://www.linutop.com/
It comes with xubuntu, 280euros.
~6W, AMD Geode LX700 433MHz, 256MB RAM, Audio, 100baseT, 4xUSB2.0

2) Zonbu Zonbox: http://linuxdevices.com/news/NS9073106297.html
It comes with Gentoo Linux, $250.
~15W, VIA C7 1.2GHz, 512MB RAM, Audio, 100baseT, 6xUSB2.0

3) Mini Linux PC: http://linuxdevices.com/news/NS6372429785.html
Not sure about which Linux flavor it comes with, but if it runs Linux, it runs Ubuntu, $99.
~5W, 200MHz x86-compatible, 128MB RAM, Audio, 100baseT, 3xUSB2.0

4) OLPC: http://www.laptop.org/laptop/hardware/specs.shtml
Not yet available, but specs are fine for a home server + external storage, ~$100.
~2W (!), AMD Geode LX-700@0.8W 433 Mhz, 256MB RAM, Audio, LinuxBIOS (!), wireless connection, 3xUSB2.0.

Many others: http://linuxdevices.com/articles/AT4923746399.html

Re:The future computing device uses less than 10W

[phantomlord]

because for a typical 300W desktop 24/7 system you probably would be paying $100/month, more than a thousand a year.

I call BS... my desktop is dual athlon 1800MP with 5 hard drives (500 watt ps)and I've also got a K6II/450 with 2 drives that acts as my personal server. Both are on 24/7. Throw in a laptop, 2 tvs (one on 12 hours a day, the other 24 hours a day for my dad), 2 waterbeds, electric water heater, 10 year old fridge, AC, etc. I used 907 kWh last month for a total of $128.64 (about 14.2 cents a kWh with all taxes, surcharges and fees). Just because my ps is rated at 500 watts doesn't mean it constantly draws that much. Even if it did, 0.3kW*24h*30d = 216 kWh/month or $30.67 a month for your hypothetical 300w desktop. You'd have to be paying over 40 cents a kWh for it to even be plausible for one computer to cost $100/month.

Re:The future computing device uses less than 10W

[stonecypher]

because for a typical 300W desktop 24/7 system you probably would be paying $100/month, more than a thousand a year.

I have no idea where you get those numbers, but they're amazingly wrong. I own an NSP. I will sell you a year of dedicated service including a ten megabit guaranteed available line for $1200/year including off-box hourly backups, and at that rate I'm making a fair profit.

If you buy five machines from me, I'll beat $1000/y. I can sell you the box, bandwidth, voltage, backups and hardware upgrades for less than you seem to believe just the wall power costs.

Stop guessing.

Re:The future computing device uses less than 10W

[stonecypher]

Sorry, that's $1200/y for a dedicated Pentium 4 3.06 with a gig of ram, an 80 gig hard drive, an IP address, any free OS (you pay extra for winders,) ten megabits of bandwidth guaranteed 99.99% available at all times and with a 99.99% uptime.

Sometimes I forget that other dedicated providers dust stuff under the rug.

Re:The future computing device uses less than 10W

[Anonymous Coward]

because for a typical 300W desktop 24/7 system you probably would be paying $100/month

Where the hell do you live that electricity costs $0.45/kWh?

Either that, or you can't do math and are just pulling figures out of a random orifice.

Re:The future computing device uses less than 10W

[tehcrazybob]

I've always been a big fan of these tiny, power-efficient computers for no reason I can fully explain. However, that doesn't excuse the tremendous errors in your reasoning.

Your first flaw is in the cost of power. It's a bit lower than that - about a fifth your estimate, in most places.

Yeah, a low-end computer these days will pretty likely have a 300-watt power supply. However, most consumer-level computers don't draw anything like that much power. Then, even if you did have a computer setup that drew 300 watts, it'd spend 90% of its time drawing around 30, with brief spikes up to its maximum for certain activities. I run a mid-range computer 24/7 on a 320-watt power supply and it costs me less than $100 per year.

If you've managed to get a computer running at full power all the time, maybe you should stop running Folding@Home, Distributed.net, SETI, or whatever it is that's using all that power. I don't have a problem with people running those programs, but you can't do that and complain about power at the same time - it's your own fault.

Re:The future computing device uses less than 10W

[scottv67]

because for a typical 300W desktop 24/7 system you probably would be paying $100/month, more than a thousand a year.

Um, I think you will want to check your math (unless you live somewhere that has verrry expensive electricity).

I ran the numbers on a Core 2 Duo E6600 box that I built a few months ago to run the Folding@Home client 24 x 7. My Killawatt says the box is consuming 155 watts when the Folding SMP client for Linux is running. The CPU is running at nearly 100%, 24 hours a day (the Linux Folding SMP client does not consume 100% of the CPU time on my Core 2 Duo - it's a software issue that the Stanford folks are working on). I should mention that the E6600 is not running at stock speed but rather is overclocked to 3.1GHz. If I went back to the factory 2.4GHz, the power consumption would decrease.

With the box running 24 hours a day, seven days a week, I calculated that it costs me $2.50 per week or $10 per month to have that machine running. Over 52 weeks, that's $130 per year to run an overclocked Core 2 Duo that is running at nearly 100% (this system does nothing but Fold all day).

So your figure of $1000 per year is a little high compared to my $130 per year. I plan to buy a Quad Core for this box next month when the Quad Core prices come down to $266. I will measure with the Killawatt again to see what the total system power consumption is with the new chip. I expect it to go up a little but it's still not going to be in the neighborhood of $1000 per year.

Re:The future computing device uses less than 10W

[redcane]

15c/kwh is normal here. (Australian Dollars). But even if it was 1 cent.... Why waste it? Some peoples annual incomes are reasonably measured in cents. Solar power (and other 100% green technologies), are still expensive to get in large quantities. If you can run 5 pcs off 100 watts, instead of one, that can be useful if your off grid. (perhaps you should be ;-)

You don't even need that

[Colin Smith]

Batch, PBS, NQS, SGE, Torque, LVS. . .

Choose your poison.

Rackable's DC solution

[blackjackshellac]

DC power is a simple way to reduce power consumption by 30%, it can also significantly reduce cooling requirements, and it's compatible with standard telco DC power kit.

dcpower

Re:Rackable's DC solution

[Firethorn]

I'd just like to point out that there is a substantial difference between 'up to 30%' as stated in the article and the '30%' quoted by blackjackshellac.

I'm not saying that there isn't potential advantages to this scheme, it's just that I wouldn't automatically assume that 30% can be saved. It all depends upon the situation. For example, if I'm using high-efficiency individual power supplies, I'm likely to save a lot less by changing over.

Re:Rackable's DC solution

[lawaetf1]

That's true in so far as the cost (power loss) of converting AC to DC as most modern power supplies are > 80% efficient. But all those individual PSUs generate a lot of heat. If you can do the conversion outside the datacenter and then run in the DC you could probably cut cooling by a third.*

*Random figure, no basis to it.

Re:Rackable's DC solution

[Firethorn]

You might be able to cut cooling to individual racks by 10-20%, but you're still going to have a massive power converter somewhere that's just as likely to need active cooling, and it's not going to be something you want to expose to the weather.

On the other hand, you probably could engineer it to be happy at higher temperatures and simply use a fan.

It's one of those things that is extremely situational dependent.

Simple solution: assembly code

[Anonymous Coward]

Google spends an absurd amount on power. If they were to write their code in assembly instead of python/java they'd see about a 10x improvement in performance, which would mean they could cut their CPU power by 10x, and thereby cut their power consumption by 10x. For many people this is not worth the extra cost of implementation, but for a company that pays so much in power one has to wonder why they haven't figured this out themselves...

Re:Simple solution: assembly code

[ChrisMaple]

Google isn't stupid; if that sort of improvement were available globally they'd be using it. My understanding of Google is that most of their work is I/O limited; assembly isn't going to help that much. Furthermore, the cost of reliably converting code to assembly would be immense and would divert labor from more profitable pursuits.

Laptop components

[Anonymous Coward]

Desktops and maybe servers should be using low power consumption laptop components for a long time.
Cutting down the desktop PCs consumption would be equally important as replacing regular light bulbs with energy efficient ones.
The technology is clearly there and mass production of low power consumption parts would result even cheaper costs.
Migrating to laptops have been started already among home users, there is no reason for desktops and low/mid-range serves to suck up as much energy as they do now.

Re:Laptop Components

[Cheeze]

i beg to differ. Most servers are NOT running 100% 24/7. If they are, they haven't been engineered properly.

Take a walk through a general use datacenter and you will find lots of 1U single use, non-clustered servers burning energy at full speed and running at .05 load. There is no real reason to run them at 300W when they can easily run at 50W with the same visible performance to the end user.

There ARE specific applications that would utilize the hardware 100%, but those are a small percentage of the servers in the real world.

Where this would become important is hundreds and thousands of servers that could run at ~50W instead of 300W. The energy savings alone to the facility would be something that would make a CFO weep.

Re:Laptop Components

[Courageous]

Most server farms are running at full speed 24 hours a day.

Afraid not. Data center utilization is typically 20%, and often a lot less. A very lot less.

C//

Re:Laptop Components

[NeverVotedBush]

Maybe at your data center. Maybe I shouldn't have said most data centers. And we aren't running underdesigned systems. We are running the fastest and biggest we can get. For modelling and simulation work, the loads are high and pretty continuous.

Re:Laptop Components

[Wesley Felter]

For modelling and simulation work, the loads are high and pretty continuous.

Sure, but this is a minority. I think it's more likely that most servers in the world are running stuff like databases, email, Web, business applications, etc. When there's no work to be done, they just sit idle.

Re:Laptop Components

[Courageous]

You're right. You should not have said "most".

Modsim is a unusual use case. That use case has its own concerns.

20% is really on the high end for utilization in virtually every data center.

The syndrome that is most alive today is the "one service, one box" issue. That's why all the drive for consolidation, coming from the virtualization vendors.

C//

Re:Laptop components

[redcane]

I think servers are probably much more important than desktops in terms of power savings, just because they are on far more of the time....

Re:Worrying text in the article:

[NeverVotedBush]

It's actually very common to have one box get elected to do any number of things - serve files, whatever. If it goes down for some reason, another takes over that function. It's redundancy and failover for reliability.

Try it

[Flying pig]

Obviously you haven't ever looked into the implications of writing a large, frequently modified program in assembler. For very low power real time systems there is no real alternative, but it is far from easy to write large programs. Why do you think compiled languages were invented in the first place? The extra cost of implementation is irrelevant when the technology just does not exist to code complex systems in assembler.

The nearest thing we have nowadays to assembler is tools for compiler optimisation and execution profiling. I expect that Google uses the latest and greatest of these. The simple fact is that nowadays Java is not particularly inefficient, especially when the competition nowadays works in more or less the same way. On the other hand it is possible to build really quite large programs out of Java relatively easily, and expect them to work. Just as you can build an apparently very efficient car (e.g. Toyota Prius) but there is a huge upfront cost in additional development and manufacturing which may outweigh savings compared to conventional technology, so the additional development cost and time to market of developing programs to use less power may exceed the savings. On the other hand, reducing processor power per cycle is a manageable hit with the huge virtue that it then has a vast field of application, i.e. almost every program that runs uses less power without further development. This is therefore the most sensible approach.

Does slowing down idle CPUs help?

[rduke15]

I'm not sure ho much power this saves, but on all servers which I install, I use this Debian HOW-TO : CPU power management page. Basically, I do:

aptitude install cpufrequtils sysfsutils
cat /proc/cpuinfo | grep "model name"
modprobe p4_clockmod ## depends on your CPU!
modprobe cpufreq_ondemand
echo ondemand > /sys/devices/system/cpu/cpu0/cpufreq/scaling_gover nor
echo p4_clockmod >>/etc/modules ## depends on your CPU!
echo cpufreq_ondemand >>/etc/modules
echo devices/system/cpu/cpu0/cpufreq/scaling_governor = ondemand >>/etc/sysctl.conf

And I see my servers run at 350MHz instead of 2.8Ghz. or more.

Of course, these are all small workgroup or very small Internet servers. It would be of no use for a server which would be at the max speed most of the time.

Anyway, I haven't had an opportunity to meter the difference yet to see how much power that really saves. Does someone know?

Re:Does slowing down idle CPUs help?

[Wesley Felter]

p4-clockmod doesn't help at all; try acpi_cpufreq. The newer Intel processors have C1E, so they automatically drop to the lowest frequency when idle, so there's not a lot for CPUfreq to do.

Re:Does slowing down idle CPUs help?

[Spoke]

To elaborate on the parent's post, p4-clockmod doesn't actually change the core clock of the CPU. All it does is force the CPU to run idle more of the time when it's in use.

p4-clockmod will actually end up causing you to use more power since it's usually more efficient to get the work done faster at a higher CPU utilization and it takes a bit of time for p4-clockmod to "ramp up" the virtual clockspeed again.

If you're running the latest kernel (2.6.21 or later) with dynticks enabled, you can install and run PowerTop which will let you identify programs and configuration issues which prevent your CPU from reaching the lowest power states.

Re:Does slowing down idle CPUs help?

[Ant P.]

Just to elaborate even more, I've tested p4-clockmod using a wattmeter on the wall socket and there's absolutely no difference between idling at 2.6GHz or 333MHz. Using powertop to get rid of useless processes actually makes it go down 2 or 3W, but since it's a P4 that's still negligible.

The only time I can see the clockmod driver being any use is when you need to force the CPU to slow down for whatever reason.

Laptop Components

[NeverVotedBush]

I've seen a number of posts about how smart it would be to use laptop components in servers. I disagree.

Most server farms are running at full speed 24 hours a day. They don't throttle back and would not spend much if any time at a low-power idle.

There are job scheduling programs where if servers aren't doing real-time stuff, they are backfilling with other jobs. Stuff gets queued up for literally weeks. It has been my experience that users demand more cycles -- not that the systems sit there idle just heating air. The systems run full speed as long as they are up. You can't idle down drives either - both for reliability (will it spin back up?), data latency, and just the fact that a data center is pretty much constantly riding a disk farm and if anything, you would want even more bandwidth than you already have.

Having them go to a low-power state when idle would really only benefit when we take the machines for software upgrades and such and we already try to minimize that time as much as possible. The bigger benefits will come from making everything as efficient as possible - i.e. power supplies, processors (and yes - even having those cut out parts that aren't being used when not used), cooling, etc.

Laptops like to idle between keystrokes and when data on drives isn't being accessed. Those stuations are really not seen in data centers.

Compatible with Life ?

[Anonymous Coward]

Someone should start a certification body for technology that is "compatible with life".

The PEM fuel cell, for instance, is life compatible, since it produces nothing toxic and no greenhouse gases (once it has been manufactured). The PEMFC bears and interesting resemblance to the cellular mitochondria -- it's sort of a very simplified form of it, which you can see if you compare the reaction cascade in the mitochondria with the reactions inside the PEMFC.

Internal combustion engines, evaporating and partially combusted fuels, road-blacktop tar fumes, auto paint solvents, and other volatile fluids used for the personal vehicles of the selfish and antisocial people who can afford to pollute are not compatible with life. Neither are coal power and very obviously nuclear energy. The exhaust and other fumes are harmful not only to our own respiratory systems, but also to micro-fauna, insects, and other life on earth (fish, acid rain). Electric trains powered by wind and solar energy are not only compatible with life, but are great places to socialize with others during your commute. Nuclear waste will be around for longer than the great pyramids have stood at the edge of a desert that was once a lush rain forest (Prior to deforestation, the silting in of a river basin, rain evaporation, soil salinization and desertification.)

More important than the number of people out there who pollute is bringing forth new technologies that are compatible with life, and taking away ones that are not. People don't like to pollute, but they are given little choice. For that reason, taking away the internal combustion engine might not be "like taking candy away from a baby" (the baby cries, the mother gave it the candy and she cripes) -- most people will welcome a transition to socially shared transportation driven by clean and renewable energy sources; those who do not are insane.

Video

[Wesley Felter]

Here's a video.

Tried it, didn't work

[Wesley Felter]

every year laptop speeds match the top speeds of the previous year's desktop machines

Comparing laptops and desktops is irrelevant when talking about data centers since they use servers. The fastest low-end server from one year ago was a 4-core 3.0 GHz Woodcrest system, but the fastest laptop today is only 2-core at 2.4 GHz. Not to mention that last year's low-end server can hold 16-32 GB of ECC RAM, and today's laptops only hold 4 GB non-ECC RAM.

RLX and HP tried building servers from laptop components back in 2001-2002, and there weren't enough customers to keep those products going.

Looking to the bigger picture

[tc9]

EnergyStar standards for power supplies are that no more than 20% of the AC power be converted to heat going through the transformer - which means that on most systems more has been turned to heat in the server room before it does anything. Just moving *that* heat conversion out of the server room where it does not need to be cooled away is a big win. Remember, it takes, in a well designed system, 1.7 times the energy to cool a space as it took energy to heat it. Opponents of DC in the server room usually miss this aspect.

Centralizing of the AC/DC conversion allows for the possibility of selecting much more efficient transformers. But we can quibble over that. In any case, isolating and concentrating just *that* portion of the heat enables new approaches. I have seen Absorption Chillers running off the transformer heat being used to boost data center cooling. I have seen domestic hot water heating. I have been in serious conversations on Stirling Engines running of the transformer heat to charge the batteries.

The highest performance versions of this are not actually about saving energy - they are about running the computer flat out. How hard can I run the machine? If the transformer and power distribution are running well, can I use my redundant power to double my machine power rather than providing redundancy? Can I do early detection of potential power failures and then back off...

Re:Looking to the bigger picture

[jbengt]

"I have seen Absorption Chillers running off the transformer heat being used to boost data center cooling"

That must be one mighty hot transformer.

Re:Looking to the bigger picture

[evilviper]

Remember, it takes, in a well designed system, 1.7 times the energy to cool a space as it took energy to heat it.

Got any source for that? It doesn't pass the laugh test.

Figures say that even the most inefficient AC units out there remove more watts of heat than they need to operate themselves.

Re:Looking to the bigger picture

[tc9]

Great - so you've got a working Carnot-Cycle generator. Cool.

Re:Looking to the bigger picture

[evilviper]

Great - so you've got a working Carnot-Cycle generator. Cool.

That's idiotic. You have absolutely no understanding of carnot.

Re:Looking to the bigger picture

[tc9]

Always happy to inform those whose mode of disccourse is hurling insults...but if you really can pull more heat out of a system than you put in energy doing so, run to the patent office...quick

From TheGreenGrid consortium

Conventional models for estimating the electrical efficiency of datacenters are grossly inaccurate for real-world installations. Currently, many manufacturers provide efficiency data for power and cooling equipment. For power equipment, efficiency is typically expressed as the percent of power out to power in. For cooling equipment, efficiency is typically expressed as the ratio of heat removed to electrical input power (coefficient of performance). Unfortunately, these individual values of efficiency often lead people to think that the efficiency losses of a datacenter can be determined by simply adding up the inefficiencies of various components.

http://www.thegreengrid.org/gg_content/Green_Grid_ Position_WP.pdf
Also from the Green Grid, estimating data center efficiency http://www.thegreengrid.org/gg_content/Green_Grid_ Guidelines_WP.pdf
Typical Numbers
IT Equipment: 30%
PDU 5%
UPS 18%
Switchgear 1%
Lighting 1%
CRAC 9%
Humidifier 3%
Chiller 33%


ASHRAE recommednations before 2005 have typically oversized data center cooling, amking the numbers worse. Ypu could look it up. Or you cou

Re:Looking to the bigger picture

[evilviper]

Your first link says nothing at all about the subject.

Ironically, a careful reading of your second link would show you how wrong your idiotic assertions are...

In their example chart on page 4 which you quote (out of order), the cooling system is responsible for 33% of energy demands. That means that while consuming 33% of the power, it is cooling the other 67%.

if you really can pull more heat out of a system than you put in energy doing so, run to the patent office...quick

Why?

Power usage is horrid

[bluefoxlucid]

Power usage is horrid. A 600MHz ARM Xscale has better performance per clock than a 600MHz x86 and eats half a watt of power; while 1.2GHz Core Solos eat at least 10 watts and most other processors around 2-3GHz eat 60-120 watts, some even 180 watts! That's between 5 and 36 times more power per clock cycle; and most of the newer chips are RISC processors on a proprietary instruction set with a real-time translator from x86 to internal RISC. It takes HOW MUCH power?

I've actually brought this up on Dell's IdeaStorm, if anyone cares (link below). Seriously, look at the OLPC, look at ARM processors, look at how modern computers work, we don't need to be burning that much power. I can't comment directly on these 100 watt video cards as of yet but does a GPU and some RAM need that much power itself? Data centers put stress on the grid, our home PCs stress the grid and cost us money, and laptops and UPS-backed machines have battery life when wall power goes away. We NEED lower power usage.

http://ideastorm.com/article/show/68285/Low_Power_ Usage_Machines

Replacing servers with a laptop

[DamonHD]

Hi,

I'm going through the exercise this month of replacing a whole slew of my always-on Internet servers at home (HTTP, SMTP, DNS, NTP) on machines going back long enough to still be running SunOS 4.1.3_U1 in one case, with a single Linux laptop. Current power consumption is ~700W. Target power consumption for the new system So, it is doable and worth doing financially, and I don't have to pay the 3x extra cost right now to remove the heat with aircon (if I did, I could pay for the solar panels too power it even in midwinter too).

See http://www.earth.org.uk/low-power-laptop.html

Rgds

Damon

Re:Worrying text in the article:

[Flying pig]

Redundancy and failover are rather different. However, if you know of a system where the servers get together and decide which one is elected to serve files (and then presumably contribute those files) I'd like to know more.

I'm not, in fact, absolutely clear what you mean in your post. There is a difference between implementing redundancy and failover as a policy, using dedicated hardware, and the idea of having servers get together and somehow vote on which is to fulfil a management function, which the article suggests.

Could you clarify what you are saying?

Replying to my stalker, the "overrated" mod

[Flying pig]

Thank you for confirming my suspicions. I wonder what I've done to upset you?

ditch the fans

[r00t]

Ideally (not stuck with hardware designed for an office environment) you do this:

Air should flow from cold aisles to hot aisles by a simple pressure difference. Those little CPU fans generate heat and lots of noise. It's better to rely on airflow supplied by the building. This of course means that the cases have ductwork and aerodynamic heat sinks as required. I've seen it for a single rack; it's really nice to eliminate the individual CPU fans. Reliability goes up (no CPU fan failures) and noise goes down.

There is some advantage to grabbing outside air, using it once, and then venting it up a chiminey. Modern computers don't need to be all that cold; for the drives it is even bad to be really cold. (see Google results) Cooling the air is expensive. Of course, some places have extreme variation in outside air temperature that must be considered.

For the UPS, water towers sound nice, but I haven't done the math. It's like having your own hydropower station.

Re:ditch the fans

[scottv67]

There is some advantage to grabbing outside air, using it once, and then venting it up a chiminey. Modern computers don't need to be all that cold; for the drives it is even bad to be really cold. (see Google results) Cooling the air is expensive. Of course, some places have extreme variation in outside air temperature that must be considered.

The flaw in that plan (pulling cold air from outside the building) is that when you bring 20 degree F air into a data center that is at 70 degrees F, you'll find the humidity in the data center getting very, very low very quickly. How much money and energy do you want to spend constantly adding humidity to the air in the data center (that nice moist air would then be exhausted to the outdoors)?

Re:ditch the fans

[Forge]

Where I come from the air coming out the back of your servers is usually cooler than the air outside. 80 degrees is average all year round. :)

Which by the way is why I have had to worry so much about cooling.

As for not going to battery. I guess we don't have any large data centres here. Our largest phone company only has around 1.6 Million subscribers. Almost twice as many clients as our largest bank.

No, But Google Has Vast Amounts of Money

[BBCWatcher]

Google isn't stupid; if that sort of improvement were available globally they'd be using it.

Google isn't stupid, which is why they aren't running the world's most power-efficient data centers. Quite the contrary, actually. My educated guess is that Lexis-Nexis has a per-search energy use profile vastly lower than Google's. (No, I don't work for Lexis-Nexis nor have any inside information. I've just reviewed public information about them.)

Google has tons of money, so they don't particularly care about energy use right now. But that lack of caring will come to and end one way or another.

Parent is right about clocking/power usage

[bigtrike]

Using a Killawatt, I was able to measure usage of some tasks with the CPU forced to 350MHz in comparison with allowing it to manage its own power. While it was by no means a well designed experiment, it seemed fairly obvious that the system always used less power when allowed to manage its own power usage.

Aero graphics, that's what....

[Joce640k]

Am I the only one who thinks Aero graphics is an environmental disaster...?

I don't know who needs 1000W but it's easy to make SLI gaming rigs go over 500W.

Stick a couple of the "twin power connector" cards in a box with a big CPU, overclock the hell out of it...that's four or five hundred watts right there.