Microsoft, Google Battle Over Energy Efficiency 164
1sockchuck writes "Microsoft and Google have opened a new front in their battle for global domination: data center energy efficiency. Just weeks after Google published data on the extreme efficiency of its previously secret data centers, Microsoft says it has achieved similar results with shipping containers (despite Google's patent) packed with up to 2,500 servers. The geeky benchmark for the battle is Power Usage Effectiveness (PUE), a green data-center metric advanced by The Green Grid. Microsoft says its containers tested at a PUE of 1.22, while Google reported an average PUE of 1.21 for its data centers, which apparently are also now using containers."
Containers (Score:5, Interesting)
Re:Containers (Score:5, Insightful)
Most businesses care about being green when it means spending less of the green ones.
Re:Containers (Score:4, Funny)
it takes a container full of servers to run Vista?
Re: (Score:2)
That's why I'd wish we'd tax the Hell out of the most non-green businesses... gotta make it worth their while somehow.
Re: (Score:3, Funny)
Joe the plumber can't afford to be green! Most small business owners making under $250,000 can't afford to be green! Won't somebody please think of the small business owners?!
Re:Containers (Score:5, Funny)
You sir, have the economic intelligence of a bullfrog.
Excellent! Bullfrogs are green!
Re: (Score:3, Interesting)
> That's why I'd wish we'd tax the Hell out of the most non-green businesses.
It should not be (called a) tax actually: it should not depend on being profitable or not. When said company pollutes, dumps etc. then it should pay for the cleanup. The only shift we need is to realize that clean water, clean air, clean soil etc. is not free.
In Europe they have a "product fee", supposed to cover the safe disposal/reuse etc. of the product at the end of its life. A step in the right direction. I would calcu
Re: (Score:2)
Then they should buy the tangerine-colored ones. I hear they're the less popular ones and so are available at huge discounts.
Re: (Score:2)
Oh, yeah. McCain is the Microsoft lobby [youtube.com].
Re: (Score:2)
Re: (Score:2)
But would a company want a building full of grubby, dinged, rusty containers? Which they'd have to modify anyway (add access panels for utilities, replace the doors with something more sensible for indoor use).
Re: (Score:2)
Well yes, no sensible company would want a rusting data center.
Re: (Score:2)
Have you ever seen an oceangoing ship? They all turn into USS Rustbucket after a few years.
Re: (Score:2)
Housing. (Score:2)
Slap a bit of paint on them; good as new!
A couple of months ago there was a story about a university using shipping containers for student housing [abc.net.au].
Re: (Score:2, Informative)
Re: (Score:2)
Power usage effectiveness isn't the whole story (Score:5, Interesting)
OK so if you have a PUE of 1.2 then five-sixths of the input energy is used to power the computer equipment. But that doesn't say how energy efficient the machines themselves are. You could be running 150W Pentium 4 Extreme Edition processors, or whatever, and still get a higher 'efficiency' than someone using Atom processors giving the same computational speed with lower power usage.
In the old days I would have suggested that Microsoft was limited to x86 processors and so they would necessarily have higher power usage than Google, who would be free to use more power-efficient architectures like ARM or PowerPC. But I get the feeling this isn't true nowadays. In servers and high-end desktops, do Intel x86 chips now offer the best bang per watt?
Re: (Score:2)
No. x86 doesn't offer the best bang per watt. Not on the hight end (IBM, ATI and NVIDIA have some nice offerings here, depending on your needs), or the low end (ARM, MIPS), or anywhere between those.
Re: (Score:2)
Do you have any performance/power figures for IBM's chips versus Intel's? I am thinking of general purpose computing, so ATI and Nvidia are out, at least until you can run Postgres, gzip, Apache, perl, Java and C# directly on a GPU core. Agreed that for number crunching and floating-point heavy applications you'd be wise not to rely on Intel's general purpose x86 chips.
Re: (Score:3, Insightful)
True - But it still means that 5/6ths of the power goes to adding computational resources rather than pure-waste overhead.
Re: (Score:2)
Exactly, also my thought is this.. If you're working super hard to get really good PUE numbers, are you really going to stop there and not think about the system itself? Getting low PUE is mostly about getting the most compute out of your watt.
People that are working hard at PUE are most likely already thinking about the watts/{flop,io,etc}.
Re: (Score:2)
In the "Old Days", NT 4.0 had build targets for X86, Alpha, Sparc, and MIPS.
Only recently did they can the other 3 architectures.
Re: (Score:2)
Idle data centers? (Score:5, Funny)
Re: (Score:3, Funny)
Why? (Score:2)
Microsoft, which is currently putting the finishing touches on a huge new data center near Chicago. The bottom floor of the $550 million facility will house at least 150 data center containers packed with servers.
So they put servers in containers, then put the containers in a warehouse? What good does the container do at that point? You're just compartmentalizing the warehouse, with really unwieldy compartments (I'll bet you can't move the containers once installed, so you're stuck with the form factor chosen at installation). Why not install modular walls instead (if it's the compartmentalization that yields the extra efficiency)?
Re: (Score:3, Funny)
Re: (Score:2)
Unwieldy for us, thinking in terms of fixing broken machines on a per-server or even a per-component basis.
For the likes of Microsoft and Google, they don't troubleshoot and repair broken HDDs, or even servers, they just roll out one whole rack and roll in another.
Changing the granularity to shipping containers just reflects the next step up - They won't bother troubleshooting
Re: (Score:2)
If they do that, they have to give up 50 % of their floor space (which is a significant factor in the cost of a data center) so they can move containers around.
Re: (Score:2)
I don't work at Google or Microsoft.
Re: (Score:2)
TFA mentions the containers are used inside a building.
Re: (Score:2)
Who says you can't put a crane in a building? As long as the ceiling is at least a bit more than twice as high (maybe 2.5x) as the containers are tall, you can put one in to pick up containers and shift them around.
Re: (Score:2)
Re: (Score:2)
You need drastically less infrastructure in the building. To build a traditional datacentre you need suspended floors, fire control systems, security, partition walls and so on. That takes a lot of time and money, time is I suspect as large a factor as money. With containers everything is built-in - power distribution, local network, cooling, fire suppression, security. You just need a wa
Re: (Score:2)
But that still leaves the question: what's the point of having something that can be moved efficiently if you're never going to move it? You're just introducing another constraint (X racks of servers have to fit within an Y-foot container) plus the expense of fitting stuff inside the container without reaping any benefit that I can see.
Think of it as a whole computing unit (Score:2)
When X% of the computers in the container are not working, you unplug it, pull it out and replace it with a container where 100% of the computers are working, and send the faulty one for servicing.
If they do that though, that means the real life PUE won't be as great. Since a fair number of containers would have nonworking machines in them.
Fat people... (Score:5, Insightful)
Re: (Score:2)
This is like two fat people drinking diet coke with their supersized double cheeseburger meal.
indeed! But it is really the guys making the burgers who ought to be audited. I can't see Google/MS having that much of a footprint - the guys that manufacture their servers, drive their containers around the world, etc., I bet, are far more environmentally costly. It would look good if Google/MS's contractors competed not only on price but also PUE. Then I think we'd see some serious savings.
Re: (Score:2)
Why? What makes you think they aren't using those CPU cycles in ways that are useful to them and their customers?
I'd be curious to hear stats on how effectively they use those cycles, but I bet they do worry about it. Most of this is motivated by cost, and unused cycles are an expensive for of inefficiency (and so likely one of the first attacked).
Re: (Score:2)
PUE is a rubbish metric for this (Score:5, Interesting)
PUE is a rubbish metric for this. The definition is nothing more than "power at utility meter" / "power used directly by IT kit". There's no account of WHAT that power is doing. Is it running one PC or a thousand? Is it hitting Gigaflops or nanoflops? You could put a laptop without a battery into a datacentre and get a PUE better than someone who has a thousand rackmounts all running at full speed. All PUE measures is the efficiency of the power conversion gear and associated equipment (e.g. UPS, etc.). In fact, UPS is an interesting measure too because the PUE of kit with a UPS would be greatly hindered in PUE stakes even against otherwise identical equipment.
Now, "Total Teraflops / Power at utility meter" - that's a more accurate metric to be comparing. And I'd guess that there Google's containers would wipe the floor with MS's (unless, of course, some trickery is being done in the TFlops measurement - you would have to carefully define what's needed). And even then, throwing a bucket load of low-power ARM processors running Linux into every square inch possible would probably thrash even Google in those stakes (unless they already do that?).
If you're going to have a contest over a metric, at least understand the metric and its shortcomings before you start claiming that X is better than Y.
Teraflops (Score:2)
Total Teraflops - that's a floating-point measurement. Not much floating point done in a database search - apart from the google rating and calculating the search speed pre haps.
Re: (Score:2)
Re: (Score:2)
That's called Data Center energy Productivity (DCeP), but you can't compare it between data centers so it's not very useful for marketing purposes.
Re: (Score:3, Interesting)
The true energy savings happen at the source. We need to find ways to increase coal-to-electricity efficiency conversion to 90% or higher.
Re: (Score:2)
At 50%, we're already getting pretty near the theoretical limit for combustion processes, iirc. I suspect you're better off finding ways not to use coal (or other fossil fuels) at all.
Re: (Score:2)
What other kind of fuel is there? (And don't say hydrogen, which is an energy sink, not a source.)
Solar has not proven to be practical. I could cover my whole roof with sun-reactive diodes, and it still won't produce enough power to run my electric heat pump during winter. Wind has the same flaw.
Re: (Score:2)
Wind, solar and hydro (including tidal) may not be complete replacements, but they are part of the solution. So is nuclear (fission and fusion). All benefit from more research, more so than the dead end that is our current fossil fuels.
Re: (Score:2)
False dilemma [wikipedia.org], anyone? Is there any reason at all that we can't do both? Does the fact that electric generation could be better make this any less of a good thing?
For the most part, data center operators can't do much to improve the efficiency of electricity generation. Reducing how much they use, however, they can control -- and a 10% reduction in coal burned per kWhr produced has the same impact as a 10% reduction in kWhrs used. And, reducing power used is probably far easier at this stage, given the
Re: (Score:2)
>>>10% reduction in coal burned per kWhr produced has the same impact as a 10% reduction in kWhrs used
Not true. The further you move down the line, the less each percent impacts the overall energy efficiency. Let's say 100 tons of coal makes 50 kilowatthours of electricity which is precisely how much energy you need to run a Microsoft container of servers (to make the math easy). Now let's suppose we take two approaches to saving energy:
- reduce power use of servers by 10%. So they use 45 kWh i
Re: (Score:2)
Your math is fine, but your conclusion is wrong. We don't care how many kilowatt hours get used, but rather how much coal gets burned (or, equivalently, how much computing gets done). Using your numbers, reducing power consumption by 10% to 45kWh means that the data center consumes 90t of coal instead of 100t -- exactly the same as if we reduce the amount of coal required to produce 50 kWh by 10% without changing the amount of power consumed.
In your credit card example, either 10% reduction applied by its
Re: (Score:2)
Power companies have been pushing their efficiencies for years - as it directly affects the bottom line. If you have any ideas, feel free to polish your resume up and submit it.
Re: (Score:2)
What about /useful/ work? If you're running N millions of instructions per second on one watt, but all but one instruction in that is Operating System Overhead.... Microsoft vs Google would report it if they knew how to measure it.
Re: (Score:2)
If you're going to have a contest over a metric, at least understand the metric and its shortcomings before you start claiming that X is better than Y.
Hear hear! It's the same with all metrics, especially environmental ones. The carbon metrics are the worst. Should we talk total carbon emitted? After all, that's what causes the problems. But the figure usually becomes meaningless. Carbon per person? Do we have an 'allowance'? Carbon per GDP? Many say it is a fix to make America look better, but if you flip it around, if we have to emit carbon anywhere more is done per unit emitted in America than anywhere else.
In short, be aware of the relevance of metric
Re: (Score:2)
I've read the article as well, PUE is about as idiotic a metric as you can have, the ratio of energy used by computer equipment versus the energy used at the facility for all other uses. The only thing that really counts is the total amount of energy used to process a given number of calculations and data requests.
You can have a really great PUE just by using the most energy inefficient computers you can get. It just seems like the googlites and M$, are just chasing each other up their own wazoos in the
Re: (Score:2)
Air conditioning costs have a much larger impact on PUE than UPS inefficiency (which ought to be very small -- do you have a reference that says otherwise?).
I agree with your general sentiment, though -- there are more interesting things to measure. Part of the problem is that different balances of CPU, disk, and network gear will produce different numbers for flops/watt, even with the same efficiency in each case.
There are plenty of places other than PUE to attack efficiency. For example, Google uses pow
Re: (Score:2)
Telling Microsoft that Google are battling? (Score:3, Insightful)
Yep (Score:3, Interesting)
It is a written rule of journalists, they economize the amount of letters in a headline. It makes sense with printed press, but at the web they should follow some different gidelines.
Re: (Score:2)
Economizing the number of letters in a headline is driven by several factors, all of which apply onscreen as they do on paper- like the need to minimize the time it takes for a reader to take in the headline, or the need to minimize the amount of acreage taken up by the headline. If anything a web based article is even more li
Re: (Score:2, Interesting)
A headline at /. starting with "Google" and "Microsoft" won't tell you much. One starting with "Energy efficiency" would be informative even when chopped. When you design headlines to be chopped, their size are much less important, that is what I was refering when I said that on the web there are different rules.
Re: (Score:2)
Geography (Score:2)
Since it is mostly irrelevant where a data center physically is, and cooling via electrical power is going to result in a comparable draw to generating the computing cycles in a warm climate, I suspect the greenest thing Google/Microsoft could do would be to site their data centers in the coldest northern climates feasible (rather than, say, California). It makes generated waste heat potentially useful as well, rather than just pumping it straight back out into the atmosphere.
(Thinking about it, Ic
Re:Geography (Score:4, Insightful)
Since it is mostly irrelevant where a data center physically is
Actually I think latency is a major issue for both Microsoft and Google as they chase the market for online applications.
Re: (Score:3, Interesting)
Since it is mostly irrelevant where a data center physically is,
well, "near a high-capacity internet link" is a pretty big issue for datacenters, and AFAIK the main reason datacenters are still being built in stupid places.
What a joke... (Score:2)
Re: (Score:2)
Re: (Score:2)
We are talking about Windows here.
They probably turn a TV on as well while they wait.
Re:What a joke... (Score:4, Informative)
Re: (Score:2)
My computer has this new green invention called a "switch". It's on the back of the computer, (so you can't hit it accidentally) and you can reduce your computer's power usage to zero while it's not in use, just by toggling the "switch".
What is considered "off" for computers is often what is termed "standby mode" by the green-conscious when referring to any other appliance.
as a recent immigrant, I notice many wall sockets here in the U.K. have a switch right on them, rather than needing to unplug a device t
Re: (Score:2)
as a recent immigrant, I notice many wall sockets here in the U.K. have a switch right on them, rather than needing to unplug a device to stop it from drawing power.
You're on 220-240 volts now, babe, it's got a lot more bite than the USA's 110 domestic power. You turn the power OFF when you're not using the outlet.
Re: (Score:3, Informative)
True, but switches on the outlet are pretty much UK-only, as are plugs that include a fuse. Other 230 V-countries don't use them.
Re: (Score:2)
100% of the 240-volt countries I'm familiar with have switched outlets. Yes, it's a sample size of two, but if Australia and Great Britain do it the rest must be merely statistical error.
Re: (Score:2)
meh, Australia's just a British colony anyway...
Re: (Score:2)
Hong Kong has the same: fuses in the plugs, and many outlets with an extra switch. Again UK influence of course.
Netherlands, Germany, France and anywhere else I have been (two dozen countries at least in total) don't have this arrangement. UK and some of it's former colonies are the exception.
Re: (Score:2)
Netherlands, Germany, France and anywhere else I have been (two dozen countries at least in total) don't have this arrangement.
Oh, well, they're sloppy on the Continent.
Doesn't everyone generalize from a single example? I know I do!
Re: (Score:2)
The US style wall socket is a ridiculously unsafe design in comparison.
I heard in the USA people actually buy stuff to cover the wall sockets to make it harder for children to stick their fingers or other stuff in them and get electrocuted.
Re: (Score:2)
Actually some people buy socket covers in the UK too, I think its a new parent thing.
The UK socket design covers the live and neutral pins The insertion of the longer Earth Pin uncovers the other two pins when the socket is in use.
Kids do like putting things in holes so a few cheap plastic covers potentially avoiding the death of your child is worth it.
Switches on wall outlets are a good thing too, since it isolates the pins from the supply and also reduces arcing on the pins. There are switchless outlets
Power usage in hibernation should be zero (Score:2)
Or have they redefined "hibernation"? Hibernation used to mean "you save all the system state to disk, and cut power". You should be able to use the big toggle switch on the back and drop AC completely.
You shouldn't need to keep a "trickle" going unless you want to use something like "wake on lan".
you have to include the PSU too (Score:3, Informative)
The PC electronics only burns 1-2 watts in standby, but the large and idle power supply will burn another 8 or so.
Or at least that's the way my imac is. I got a watt meter and it's 70w at full power, 40w in low-power mode, 10w in standby and 10w when off. It only goes to zero when you unplug it.
My laptop is the same: the charger burns 7w even when you don't plug it in to the laptop.
Re: (Score:2)
I think you missed the point. From your very own comment it sounds like what they were saying is that if you have your PC turned off for 16hrs of the day using 1 watt and being able to turn it on and be productive instantly is better than sitting waiting for a couple of minutes using 100s of watts for the system to boot up from full power off before you can be productive. It sounds like what they were saying is effectively that a few minutes of time where you can't do anything at 100 watts is worse than hou
Re: (Score:3, Informative)
A "typical" PC, of which there are none, will likely pull 125-200W at startup. It runs full out, afaik, until power management kicks in. For my laptop, it takes nearly 5 minutes* from power switch to useful (as judged by both disc activity and inability to accept keystrokes in realtime). So 1/12 hr x 125W = 10 watt-hours. That's ten hours in standby if standby is 1W over hibernation/off.
It has a huge benefit to usability, though. Being able to "turn on" the machine and have a working browser over a wireless
Re: (Score:2)
Well, call me insane, but I don't think that some 500kw (how big is the US population?) are a big amount of power for an entire country.
SWaP (Score:2, Informative)
I like Sun's SWaP metric [sun.com] because its value is based on a business operation that you can define.
And as the article mentions, datacentres in a shipping container are like, sooo 2006 [sun.com] .
Re: (Score:3, Insightful)
"despite Google's patent" (Score:2, Informative)
You do know that a patent doesn't prevent you from building and using a patented device? You just can't sell them. In fact, making the information available was the reason for patents.
Re: (Score:2)
"Green Grid" has no Green Organizations as Members (Score:3, Insightful)
The Green Grid: Members List [thegreengrid.org]
Go straight to the source. (Score:2)
The website linked to basically regurgitates material from a Google website about their data centres [google.com] and a blog entry [wordpress.com] by a Microsoft data centre employee.
The original links are more informative than the rehash.
Tagged "pueispants"? (Score:2)
How do containers help improve efficiency? (Score:2)
That isn't answered by TFA, and I don't see how packing a few racks of servers in a large metal box would help make a datacenter more efficient.
Re: (Score:2)
It doesn't.
Unless you can somehow benefit from being able to quickly deploy a datacenter somewhere, the only benefit of a container is that you can create a permanent datacenter in the parking lot. Then you won't have to pay rent for the floor space and can spend years promising your neighbors and building's owners to remove those damn containers really soon now.
Google brags of using evaporative cooling (Score:2)
To save power over phase-change (i.e. freon-style) cooling (and yes, I do realize evaporation is a phase-change, I didn't pick the name).
So they save on electricity and instead use a lot more water. Is this a big advance? Is the energy cost of getting that water there counted in the efficiency rating?
If Google really wants to reduce the energy used (and not just their electrical bills), they need to look at these kinds of things.
Obligatory "Murder by Death" quote (paraphrase) (Score:2)
"Microsoft, Google Battle Over Energy Efficiency"
And! Microsoft and Google! Say your goddamn conjunctions!
Beowulf cluster anyone? (Score:2)
I'd like to have a Beowulf cluster of these shipping containers...
Or better yet, imagine a container ship filled with them--how many MIPS is that, anyway?
(Dyson sphere [wikipedia.org]--here we come!)
P.S. Never underestimate the bandwidth of a Panamax freighter filled with Flash chips...
Damned lies and ... (Score:2)
On the other hand I fully agree with the other comments t
Re: (Score:2)
If you're in the USA, you'd be getting maybe 600 watts of sunlight per square meter, so that's 9kw of sunlight.
Assuming 40% efficient solar panels that means 3.6 kilowatts. Assuming each server uses 100-150W. That's enough power for about 25 machines, not inclusive of cooling.
Cooling is the tricky bit. An airconditioner could easily use 1 kilowatt.
But 1 kilowatt can power a lot of fans. So if ambient air is cool enough, you might skip the hea
Re: (Score:2)
I'm quite north enough already [google.ca], and I'm using my computers to heat my apartment right now (it's 5 degrees celsius outside).
P.S.: for a quick laugh, zoom out of the map a bit. That's what is called "living in the middle of nowhere". Or at least it sure feels like it.
Re: (Score:2)
Or a threesome...
iMicroogle?