Inside the Google-Plex

tappytibbins writes "Baseline magazine has an in-depth story about how Google manages its own IT infrastructure. From the article: 'In general, Google has a split personality when it comes to questions about its back-end systems. To the media, its answer is, "Sorry, we don't talk about our infrastructure." Yet, Google engineers crack the door open wider when addressing computer science audiences, such as rooms full of graduate students whom it is interested in recruiting.'"

print friendly version

print friendly version [baselinemag.com], because their page layout is a little too far on the "hey, if we add more adverts, we'll make more money!!1!"/WiReD-more-colors-are-good end of the scale.

Elevator Pitch Version:

It's not the highly parallel clustered racks of custom-designed linux servers that makes Google Google. That's an enabling feature. Rather, it's their custom engineered application-level operating environment, if you will, which runs on top of that. It's good at keeping data, indexing attributes, finding it, breaking problems down, and intelligently routing work and results. The search engine and all their other apps are built on top of this, and it allows their engineers to leverage this common distributed platform in all of their external and internal applications.

=== End Elevator Summary ===

Not many companies are willing to write their own application layers to deploy services. Most companies CAN'T. It's just not worth it. It's worth it to Google because developing and deploying world-wide information retrieval services is their business.

However, a standardized Application OE that can run and take advantages of the resources of many potentially unreliable computing resources would be very valuable to many businesses.
Grid technologies, web services, J2EE, and clustering technologies are just scratching the surface.

Forget the googleplex

I want inside the google party plane!!

Also...

On a related note: "Behind the Glass Curtain: [metropolismag.com] Google's new headquarters balances its utopian desire for transparency with its very real need for privacy."

I'm still waiting for pictures of the "party plane", though.

Re:Also...

Isn't PageRank based on a voting system? OK, in an intranet full of PDFs and Word docs, who votes? Yup, that's right, PageRank doesn't work in an enterprise context.

I think it is fairly obvious that Google does not use PageRank for some of its specialized searches, such as for gmail, Google video and so on; they simply use some text-matching algorithm in those apps. Not that hard to imagine that this would be the case with Google Enterprise (or whatever they call their app).

Come to think of it, why do they need that many? Yahoo, MSN, Altavista, and everyone else index the web with far fewer I'll bet

Unless you're trolling, (or think Google should be a search engine alone), you would know that Google does a lot more than plain-vanilla web search.

Don't get me started. 5,000 PhDs and their efforts to combat Click Fraud amount to -erm- not a whole hill o'beans.

Try reading that ZDNet article, instead of forming opinions just by looking at headlines.

Friendly

Its nice to know that some companies are willing to open their doors to the Tech comunity. Reminds me of Open Source Software... but only with hardware

It still worries me that google will soon know everything about everyone. I hope they dont share that data with ANYONE.

450,000 servers?

I feel bad for the sysadmin

Re:Sysasmin(S)

I'm sure they need more than that. Google representatives often say (when talking about cheap commodity hardware), "With 1000 machines, you can expect one to fail everyday." Therefore, if they have 450,000 machines, you can expect about 450 to fail a day. Not only that, but they are probably adding machines like crazy and replacing old machines as they become cost-inefficient (the numbers I've heard say they keep computers for 2-3 years). I think it would take more than two guys to do all that. I'm sure they have a huge ratio of computers to sysadmins, but they still need a bunch of folks to replace the dead machines and add new ones. I imagine their servers are easy to manage on the software side, however.

Re:Sysasmin(S)

i seem to remember reading they don't replace the failed ones, they just junk the entire rack when it becomes not worth running anymore.

i'm guessing google are big enough to have thier own datacenters and thus not have space at such a premium as smaller operations. If space isn't at a premium replacing a machine in a rack probablly isn't worth it (it means you have a machine whose remaining usefull life is out of sync with the rest of the rack its in).

NDA time?

Looks like it's time for Google to start making prospective students sign non-disclosure agreements...

Re:NDA time?

Aha! He has revealed that the Google-plex contains a 'door.'

This slip of the mind will prove invaluable in my Google-imitation plots.

That makes perfect sense

Most journalists and business analysts are notable for doing a half-assed job and taking credit for cut & paste jobs. Journalists who actually spend time researching their stories are a dying breed, so my take on this is that Google would rather not waste their time answering stupid questions from people who don't even understand what they're publishing. Their time is much better spent recruiting smart people or just talking to grad students in some sort of academic goodwill.

A Google Lecture Experience

Google engineers crack the door open wider when addressing computer science audiences, such as rooms full of graduate students whom it is interested in recruiting

An alum of my university who works at Google recently visited and gave an informative lecture with a long Q&A session. I can vouch for the fact that we were told more than I've ever been able to read online about the way Google manages various issues, like their IT infrastucture. However there were still limitations to what he would/could tell us (sorry I won't go into specifics). It seemed (as you would expect) the better our questions, the better his answers, and if we asked questions that were too good, then it was likely that he did not feel liberated to answer.

Also, Google was cool enough to sponsor a Programming Contest and a Graduate Research Conference we held. Our alum attended our little conference and had great feedback and questions for our presenting students. With respect to knowledge, intelligence, and humor this guy was all I would imagine and/or hope for one of our alums working at Google.

On the otherhand, I was very unimpressed with certain issues concerning lack of professionalism in the lecture. As one example, though this is only an impression, it seemed that he felt he could just get away with wearing jeans and a Google t-shirt for the few days that he was with us because he worked at the ever prestigious Google. It seemed a bit arrogant. Also keep in mind that his position at google is higher than a solutions engineer.

Just thought I'd share.

Re:A Google Lecture Experience

On the otherhand, I was very unimpressed with certain issues concerning lack of professionalism in the lecture. As one example, though this is only an impression, it seemed that he felt he could just get away with wearing jeans and a Google t-shirt for the few days that he was with us because he worked at the ever prestigious Google. It seemed a bit arrogant.

Why does he have to wear anything more than jeans & a Google shirt? A computer guy in a shirt & tie is not a happy computer guy.

It's been my experience that the companies who worried most about what their IT staff was wearing were the worst to work for.

Re:A Google Lecture Experience

Fair point, but perhaps people who equate clothes with professionalism (not accusing you, and it is true that we all make judgements on appearance to some degree, but anyway...) have been deemed as people who they do not feel the need to impress. If I owned a business I would certainly be inclined to think that way - maybe that's a reason that I wouldn't succeed in owning a business, but it appears that Google makes as much money as they could ever need; if they don't become too greedy they can afford to interact with the world on their own terms rather than those decided by a particular portion of society.

In essence, I see the jeans and a T-shirt as a positive representation of the company, it represents a company in which essentially arbitrary rules are examined for their usefulness and if neccessary discarded.

Re:A Google Lecture Experience

A few decades ago in a small service bureau near LAX a visiting suit-and-tie type walked past and took exception to the engineer who was dressed in shirt and jeans and busily filing down a resistor on a logic card. Seems he wanted the miscreant fired/evicted for (a) not representing CDC to the public correctly, (b) obviously damaging a very rare and expensive piece of equipment, and (c) ignoring him. "That's Seymour," said the DC manager, "and I don't think you can fire him. And if he's filing down a resistor, I presume the computer will work better that way".

Please accept the above for the lovely second-hand urban myth that it is, one belonging to a CDC 6600 site where I was lucky enough to attend a few lectures.

Re:On the other hand...

That's an interesting point. I think a lot of companies are actually that way. I work for *undisclosed faceless corporation* and people show up here in shorts and birkenstocks. A guy on our team walks around the offices barefoot. Invader Zim posters, figurines, calendars of cheerleaders, etc. are all over the place. You could show up in flip flops if you wanted to... but people choose not to. There's something about that Google mentality that sounds neat at first, but then you realize that you're not in college anymore, and even though you CAN wear flip flops to work, you probably don't want to.

So is it neat to have a trendy office space? Sure. Is it neat to have communal centres scattered around the building, and be encouraged to stay afterhours to play games? I guess. But it's the kind of thing that gets old once you realize you've got a family and a life outside of work. Working for Google sounds like working in a basement with a bunch of friends, but that only really works if you don't have other things you want to be devoting time to. Once their workforce matures a bit, I'd guess their "kooky, trippy workspace" won't work quite so well. Don't forget, they're still basically a glorified startup. I'm sure Microsoft had a lot of the same feel back in '86.

Re:A Google Lecture Experience

On the otherhand, I was very unimpressed with certain issues concerning lack of professionalism in the lecture. As one example, though this is only an impression, it seemed that he felt he could just get away with wearing jeans and a Google t-shirt for the few days that he was with us because he worked at the ever prestigious Google. It seemed a bit arrogant.

Shock. Horror. The man was dressed comfortably in clean, casual clothes. That's your only complaint? Where are you from, Victorian England? Come join the rest of us in the 21st century. Arrogant? Arrogance would be whipping his thingy out and pissing on the front row. For casual dress to be arrogance it would be necessary for him to believe that a suit and tie is the only appropriate dress for such an occasion, and then intentionally not wear them as an insult. I doubt this was the case. Sounds to me more like the problem's with you and your slavish adherence to an increasingly outdated dress code.

Re:A Google Lecture Experience

Perhaps "arrogant" wasn't the right word for the poster to use, but I can see where he's coming from with this. If you're presenting at a prestigious institution (or one that considers itself such) you should show a little respect for that institution by dressing it up a bit. A tie wouldn't be necessary, but ditching the jeans and wearing a button up shirt (or a polo shirt even) seems like it would be in order.

I'm a computer guy, I wear jeans to work every day, and I'm happy that way. If I'm being asked to speak in front of a large group of people I don't grab something out of the hamper, y'know?

Post-Beta

From TFA:

On the other hand, Google doesn't hesitate to create applications for internal use and put them on its own server grid. If the company's software engineers think they can tinker up something to make themselves more productive--for example, a custom-built project tracking system--Merrill doesn't stand in their way.

If it is anything like their web-presence, half the stuff must have 'Beta' appended to it.

New GPayRoll-Beta!

I once had an interest in Google

I was interested in working for Google -- mostly for a job. I even had someone from inside recommend me -- I figured I'd be in, no problem. Rather than being interviewed for my skills, or the relevant department, they interviewed me for a sysadmin. No problem, I'm a sysadmin I thought. I didn't do the CS route at a university, though, and there were some highly relevant things that I just didn't know how to answer. I didn't pass round two.

So I tried to get another interview for a while, but no bites. Google has made it clear that they aren't interested in my work. I've stopped trying to get a job there. Besides, I don't think I can sit in a cube and take long drudgery with the occasional stinging bits of punnishment. I like all my punnishment unending and all at once, and so I just work for startups.

Re:I once had an interest in Google

What background do you have in software development? Currently Google interviews people for "core basics", which are the basic skills you would learn by going to university or trade school in the field related to your position. For example, when interviewing for the Java Developer Software Engineering position, you'll get a lot of questions about the collections library, synchronization, and core computer science questions like semaphores and two-phase commit. My experience with Java, and I know I'd completely fail a system administration interview.

Perhaps you should have informed your recruiter about your background?

Note the context

> "Sorry, we don't talk about our infrastructure." Yet, Google engineers crack the door open wider when addressing computer science audiences, such as rooms full of graduate students whom it is interested in recruiting.'"

So the question to ask is: Who is `we'? I could provide an answer. `We' here, is Google's official answer to such questions from the ever inquisitive press men. Those who speak on behalf of Google have been asked to memorize that answer if they do not wish to talk about a topic. It works well. This approach reminds me of the government's `We can neither confirm nor deny...' mantra.

What follows next could be interpreted as a Google engineer's answer to a question, which answer may simply represent one of many possible implementations and NOT one you could find supporting Googles infrastructure.

Guys, this is all about semantics and context. Good night!

In other news

In other news, Google has started tagging new employees on probation with 'BETA' labels.

Little in the way of structuring data

Everyone is talking about GoogleFS. But no one is talking about how they manage structured data. How do they do it? Some SQL stuff, some homegrow potion, or have they managed to create a sensible interface for structured data on top of GoogleFS?

Just another Google Publicity Stunt

This is probably just another Google Publicity Stunt. I've worked at several of the top companies, and its just "people" that work there, not much more "intelligent" or better than others who choose a lower profile, more fun and probably more lucrative environment to work in. Who has any information to prove it otherwise?

mb

go to the source

Here are some good papers [google.com] about Google's technologies:
Sawzall [google.com] (simplified scripting on top of MapReduce)
MapReduce [google.com] (Google's massively parallel system based on the concept found in functional programming. The system takes care of managing jobs, parallelism, and fault tolerance, allowing engineers to more quickly produce code.)
GFS [google.com] (Google's File System)
Google's Cluster [google.com] (An older paper describing how Google's search cluster works. The cluster described in this paper is a few generations out of date.)
BigTable [andrewhitchcock.org] (Google's semi-structured database. There haven't been any papers released, but this is my write up based on a talk given in October 2005.)

And here are some videos:
The Google Linux Cluster [washington.edu]. This is an older video where Urs Hoelzle talks about their system and focuses more on the hardware side of things.
Google: A Behind-the-scenes Look [washington.edu]. Jeff Dean gives an overview of most of the technologies mentioned in papers above. I thought the demonstration of Google's internal word clustering was interesting (and funny).
Perspectives on the Information Industry [washington.edu]. This is a technology-light (IIRC) talk given by Eric Schmidt.
BigTable: A Distributed Structured Storage System [washington.edu]. The talk from which I created my BigTables notes (above).

Andrew

Ah, I thought it would be something like this

An Inside Look at Google [google.com]

Hey guys, watch it, it's about the women at Google! ;-)

I guess now we have yet another reason to go there. :-p

My Google Recruitment experience

I went through the majority of the process (phone interviews, then the face time at the New York facility and a trip to California) before I told them I wasn't interested in the job. My reasons for turning them down were three. First, the 80/20 deal was a myth if you were going to pursue something they were really interested in. I wanted to complete my PhD. Now, this PhD was not in a field they cared much about (despite their glaring need of my skills for one particular service), so that part of the deal was mumbled everytime it was mention. Second, the pay wasn't good enough to basically live at the facility. Third, the interview process was abusive in many respects. The first phone call was with a guy consumed by asking me about my doctoral research and my knowledge of how inodes work. He kept shifting between the two. When I asked him why this was even necessary given the position I was applying for, he got irritated and said that you had to know the ins-and-outs of how a file system works in order to configure (something I wouldn't be doing) any part of their infrastructure.

This lead to my observation of part of their file storage system which is quite possibly the most tweaked NFS nightmare/genius/what-the-fuck I'd ever seen. My past experience with networked file system was, I admit, very limited compared to what they had going on. Now, again, I wasn't even going to have anything to do with this system or any sysadmin work at all, but it was obvious that they wanted you to at least have knowledge of the system on some level beyond the user. It also came across as a showing-off culture too. I am glad I didn't take the job for various reasons, but if you are a sysadmin freaker who loves dinking with shit, you'd fit in; especially if you like to show it off too. Just be prepared to have some middle manager there fuck with you for a hour or two on the phone before you get to the outer part of the inner sanctum.

GoogleFS and You

Little in the way of structuring data
by leandrod (17766)

Everyone is talking about GoogleFS. But no one is talking about how they manage structured data. How do they do it? Some SQL stuff, some homegrow potion, or have they managed to create a sensible interface for structured data on top of GoogleFS?
--
Leandro Guimarães Faria Corcete DUTRA
DA, DBA, SysAdmin, Data Modeller
GNU Project, Debian GNU/Linux

----------8<----------

Wild Guess&#174; ?
Rob Pike?

http://herpolhode.com/rob/

http://plan9.bell-labs.com/plan9/

http://plan9.bell-labs.com/wiki/plan9/Papers/index .html

http://plan9.bell-labs.com/sys/doc/

http://www.vitanuova.com/inferno/

http://herpolhode.com/rob/hotchips.html

leandrod,
You might find this especially interesting considering your affiliation.

GPL?

They sell Google-in-a-box appliances without root access or source. If it contains Red Hat Linux isn't it a violation of GPL?

Similarities with the Borg cube?

What are Googleplex's similarities with the Borg cube? does it repair itself when a node goes down? is it possible to bring the whole system down by injecting a false command in a node? etc

A whole lot of words about nothing?

Am I the only one who got the feeling that the entire article was one big babble about things everybody who reads any given daily newspapers knows? Amazing how one can craete an entire article that says absolutely nothing.

Romance?

"With his unruly hair dipping across his forehead, Douglas Merrill walks up to the lectern set up in a ballroom of the Arizona Biltmore Resort and Spa, looking like a slightly rumpled university professor about to start a lecture."

This sounds more like the beginning of some Harlequin romance novel than the first line of a serious piece of journalism. The sentence even manages to run-on and has a simile. I expect that by the end of the article his hairy chest will be bared by the protagonist and that they will live happily ever after.

Re:It's hard to believe...

You think hundreds of machines is ridiculous...

I read that *many* of those machines have more than 640K of RAM!

Re:It's hard to believe...

...that a single search engine runs on hundreds of servers. What do they need all of those machines for?

Probably for all of the web pages they cache [google.com].

Re:It's hard to believe...

They need them for serving news, mail, maps, satellite imagery, images (which include cached thumbnails), video, and of course, text ads.

this list is far from complete [google.com]

Re:It's hard to believe...

..that a single search engine runs on hundreds of servers. What do they need all of those machines for?

you dropped a few zeros there