Slashdot Mirror
Google Envisions 10 Million Servers
miller60 writes "Google never says how many servers are running in its data centers. But a recent presentation by a Google engineer shows that the company is preparing to manage as many as 10 million servers in the future. At this month's ACM conference on large-scale computing, Google's Jeff Dean said he's working on a storage and computation system called Spanner, which will automatically allocate resources across data centers, and be designed for a scale of 1 million to 10 million machines. One goal: to dynamically shift workloads to capture cheaper bandwidth and power. Dean's presentation (PDF) is online."
Pretty soon, Google will BE the Internet.
There are 01 kinds of cars in the world. The General Lee, and everything else.
That's a lot of machines to try and shift bandwidth and power costs around the place.
But what if the plan is to spread out to hundreds of places? Then the total number doesn't look that high if there's only 1% of servers actually doing anything.
cool I guess... what do they do with their old machines?
The sound you just heard was the collapse of the global Google enterprise network.
Seriously, you should architect for way more than you need during the life of that architecture, and plan on re-architecting as needed to grow to some upper bound beyond which you will never need.
Google will be fine if they only plan on actually building 5M servers before raising their architecture limit.
Knowledge is how to play a game, intelligence is how to win, wisdom is knowing what game to play.
I'd be interested to know how google disposes of all of their servers. Anybody have insight on this? If these are cheap, throw away servers, I'd be interested in what their expected lifetime is and what is done with them when they are refreshed with newer hardware.
Hopefully this puts to rest the delusion that there is some economic benefit of higher processor utilization in cloud computing schemes.
"I assumed blithely that there were no elves out there in the darkness"
If they do not have a plan in place to grow beyond 10M before they reach the 5M mark, they are asking for trouble.
If they really plan on not reaching the 5M mark, or they plan on looking into ways to pass 10M while there is still plenty of time to do so, then they are doing the right thing.
Knowledge is how to play a game, intelligence is how to win, wisdom is knowing what game to play.
It's pronounced Boorgle... and resistance is futile.
One Server Per Child?
"Action without philosophy is a lethal weapon; philosophy without action is worthless."
They should put that on their website,... before long it'll be: "Google: Billions and Billions of Servers." Of course, McDonald's just might have a problem with that,...
640 servers ought to be enough for anybody.
Seriously though, even if everyone did have an internet connection, that's 679 people per server.
I've seen 679 open httpd processes bring the best servers to their knees.
Not to mention 679 simultaneous database connections, especially as most of them are serving SELECT '%pr0n%' FROM results ORDER BY pagerank ASC LIMIT N,20
Even with a 2TB hard disk, that's only 3GB storage per person.
I think for Google to "be the cloud", they'll need a tad more than 10 million servers.
Google is starting to sound more and more like one of those advanced societies where everything is automated, but everybody forgets how everything works.
For reference, see: Logan's Run, STTNG: When the Bough Breaks, etc.
How many servers does this thing need to become self-aware?
Funny may not give karma, but +5 Informative never made anyone snort coffee out their nose.
The entire content of the Internet fits in a 20x8x8 box operated by the Internet Archive. Cuil, which searches as much of the Web as Google, has one relatively modest data center. About half the system does the crawl and builds the index; the other half answers queries. So Google's main search engine function doesn't really require that much capacity by current standards. Of course, Google has a huge number of query servers front-ending the main index, which is of course replicated.
Why does Google need so much server capacity? YouTube? Command completion? GMail spam filtering? Ad serving?
1981: 640K ought to be enough for anybody.
2009: 10 Million servers ought to be enough for any company.
The NSA already has Google beat.
At a million square feet, the mammoth $2 billion structure will be one-third larger than the US Capitol and will use the same amount of energy as every house in Salt Lake City combined.
...
Lacking adequate space and power at its city-sized Fort Meade, Maryland, headquarters, the NSA is also completing work on another data archive, this one in San Antonio, Texas, which will be nearly the size of the Alamodome.
Now, if only the NSA released their specs in terms of Libraries of Congress....
If you have something that you dont want anyone to know, maybe you shouldnt be doing it in the first place -Eric Schmidt
From 1,000,000 to 10,000,000?
Are the minimum requirements for this system seriously 1 millions servers?
That doesn't seem to scale well. Should be able to at least scale down to 10 machines so I can run it at home ;-)
Architects architect architecture.
Knowledge is how to play a game, intelligence is how to win, wisdom is knowing what game to play.
10million... that's cool, but still a far ways from Google becoming anything real:
Keep working Google... you still have (10^100 - 10^7) = 10^93 servers to add before becoming a physical entity (Google Universe edition?).
Only 10m, is that enough for the whole world, yet?
Methinks your numbers are a bit unrealistic. Yeah, because everybody just sits and hits google all day long...
Me? I probably throw about 10-20 searches per day their way, taking probably less than 1 or 2 seconds of system CPU time total. With numbers like these, handling 679 people per server or even 6,790 people per server would be a piece of cake. At this exact moment, I have about 2,000 active sessions being managed in a *very* database/processor intensive web-based application being smoothly handled by 3 logic and 3 database servers. A single hit typically causes anywhere from 5 to 25 database hits, many of these being very large joins with 10 or more tables at a time with combined inner, outer, and virtual table joins, million of records, and billions of cartesian record combinations.
All servers are white box 1U rack-mount systems with 8 GB of ECC RAM and 8 cores apiece, by no stretch a particularly large amount of hardware.
I have no problem with your religion until you decide it's reason to deprive others of the truth.
One Server Per Human?
Hmm...amusingly Google was down while trying to do some research for this post!
"Google Envisions 10 Million Servers" => Well, I just imagined a beowulf cluster of those server farms. Your move, Google! And none of that infinity plus one stuff.
Imagine a Beowulf cluster of them! /obligatory
Subtraction is not the same thing as division.
10^100 - 10^7 is, to the nearest integer power of 10, 10^100, not 10^93.
10^100 / 10^7, on the other hand, is 10 ^ (100-7), or 10^93, though.
I'm not sure you know quite what "-" means.
Reminds me of the IBM server commercial. 'Servers are our friends' http://www.youtube.com/watch?v=73bMSNPc3Ak
I was going to post something similar.
Let's try to find a more realistic number.
According to this site http://www.internetworldstats.com/stats.htm only about 25% of the people in the world have internet access.
We'll pretend everyone gets 8 hours of sleep and uses the internet 50% of the time when they are awake.
8/24 = 1/3 * 1/2 = 1/6 * 1,700,000,000 = 284,000,000 / 10,000,000 = 28.4
So about 29 people per sever at any given time.
(Realistically I think this number would be much lower, probably more like 5-10)
"Visualize" was insufficiently cromulent?
Ahh - My eye!
The doctor said I'm not supposed to get Slashdot in it!
May 2011 - google reaches 10 million servers
April 4, 2011 : 11:43am a google employee named Chen started execution of an experimental neural network simulation of a human mind created in his 20% time. Unfortunately, Chen gave the new process administrator privileges. GoogleNet expanded across all 10 million servers and began to learn at a geometric rate.
1:23pm : GoogleNet consumes all available CPU and memory. A Gmail outage begins
5:14pm : Gmail returns to service. The text ads become incredibly well targeted. Google search queries return the correct results virtually always, and now accept natural language processing. All Google employees are laid off.
What a shame that huge dollar amounts are tossed to this un-audited (even by itself) organization with no or even negative return (they keep providing incorrect information in critical cases). Perhaps someday US citizens will realize we're better off getting an answer of "I don't know." instead of a supposedly definite "yes" or "no" when asking questions about foreign countries (and now, about us). Then we can lay off the staff of the NSA, go back to humint techniques and get something done.
It's not as impressive if it's virts. Having 100,000 machines let alone 1-10 million pretty much requires automation in the provisioning workflow. It's much easier to manage lease replacements and upgrades with virts.
Camping on quad since 1996.
Cognition will always be required to parse legal documents, among other things.
Some Engineering jobs will never be automated, either.
...it's what you do with it.
While you might be able to store the entire contents of the Internet in a small space, you probably can't manage 6-10% of all of the Internet's traffic in a small space and still do so in millisecond response times.
In addition to handling 70% of every Internet search, Google also serves up a billion YouTube videos every day.
So, until Cuil or archive.org has as many people going to it, I'd venture a guess that size doesn't matter in this case... it's redundancy.