Google Two Years Into Overhaul of the Google File System

El Reg writes "As its ten-year-old file system — GFS — struggles to keep up with Gmail, YouTube, and other apps it was never designed to support, Google is brewing a replacement. According to the company, it's two years into a GFS sequel designed specifically for customer-facing apps that require ultra low latency."

Comments

hmm

[gnarfel]

Well I'm no expert on Google's internal workings, but are any of these protocols or file systems they've developed been released outside of Google for public use?

Re:hmm

[buchner.johannes]

GFS is proprietary and for internal use only. The only released a paper describing how it works (don't know if that content is enough to rebuild it). I think GFS (global file system) from Redhat and OpenGFS is something differently. Hadoop is what you want. What would we do without the wiki

Re:hmm

[Brian Gordon]

No, they haven't. So why does the editor think we care? "Google Six Months Into Resurfacing Parking Lot"

Re:hmm

[mysidia]

They have not, and apparently Google thinks of the Google FS as part of their secret sauce, such that they will probably never get it released. Although they seem happy to write papers about it.

It's actually really sad... Google has built an innovative platform for distributed computing, that solves quite a few problems, vastly superior to the state of the art in distributed computing, but they basically keep the filesystem and clustering implementations completely to themselves, it would seem.

They use the Linux platform to the absolute max, leveraging all the blood and sweat Linux developers poured into its development over the past 15 years, and yet, not contributing back any of their most significant enhancements.

I won't call it evil, as they're under no obligation to release GoogleFS or their map reduce implementations, it's just unkind.

I would equate it to an inventor creating the lightbulb, and their employer saw this, and decided instead of trying to sell the invention to the public, they decided to only allow their own factories to buy lightbulbs, thus netting them a competitive advantage over other factories whose workers had to operate in the dark or by candlelight.

No software product available to the public that even utilizes GoogleFS. Instead it's all software as a service (The Google search engine service, that is)

Re:hmm

[ToadProphet]

Parent and GP modded funny? Am I missing the joke or are there some giddy drunks with mod points?

Re:hmm

[buchner.johannes]

Everyone is in a good mood. Why not :-)

Re:hmm

[MeatBag PussRocket]

They use the Linux platform to the absolute max, leveraging all the blood and sweat Linux developers poured into its development over the past 15 years, and yet, not contributing back any of their most significant enhancements.

i see your point, but its not like google isnt giving signifigantly in return. most people would be hard pressed to deny that Googles search engine was a game changer in the interweb. at its release it was leaps and bounds better tahn just about anything out there, and is still the gold standard for finding information. hell they gave us the verb "to google" we got a pretty decent browser out of it, gmail, google docs, google maps, and a whole bunch of other stuff they've generated. not to mention a forthcoming OS. at this point i can already hear critics screaming about Googles profits driving these services, and you know what, maybe they are, but i havent paid Google a dime, and most likely, neither have you. i dont care if they make money, theres nothing wrong with it, and i'm even happier that they make money without involving me whatsoever. in many ways i would think Google would be a champion to the FOSS community. so they want to keep a filesystem proprietary, frankly thats not so bad, competition is good but competitors arent usually. Google is a good counter balance to Microsoft and other would-be owners of the interwebs. are they "good" as in saintly? no, but they never claimed to be, they claimed "dont be evil" i'd say they're pretty far from that.

Re:hmm

[ksatyr]

"Google Six Months Into Resurfacing Parking Lot"

And it's still in beta.

Re:hmm

[lawpoop]

They use the Linux platform to the absolute max, leveraging all the blood and sweat Linux developers poured into its development over the past 15 years, and yet, not contributing back any of their most significant enhancements.

Not contributing back!? Dude, they gave us *google*. Remember what it was like before google? When internet search was basically voo-doo crapshoots, that worked 25% of the time? They gave us a search engine that actually *worked*. Before that, you basically had to bookmark or memorize internet sites that you liked. Good luck actually finding what you were looking for without having an actual site in mind beforehand.

I think that alone has probably spurred the development of free software. Imagine being able to *find things* on the internet!

Re:hmm

[Night Goat]

Yahoo worked fine for me before Google. I think you give it more credit than it deserves. The downside of Yahoo was its advertising and clutter. The searching part worked fine.

Re:hmm

[Snarf You]

While I found both posts informative, I find it funny that they were modded funny. It's meta-funny. You know what else is funny, is that the word funny starts to sound funny after saying it enough times.

Re:hmm

[billcopc]

You clearly weren't an Altavista user.

Google's results today are no better than the leading search engines 10 years ago. People were gaming the engines then, and Google came up with a smarter algorithm (Pagerank), but today's results page is again full of garbage because people learned how to game Pagerank. Combine that with the web 2.0 fad of scraping and regurgitating everyone else's content, and the resultant pile of URLs for any given keyword is utterly worthless. I call it "metapublishing", because the content is worthless, it's become a twisted game of outwitting Google to maximize ad revenue while providing zero value.

Searching has always been a game of finding the most specific yet least popular terms to define what you want, and then adding a bunch of negative keywords to filter out the junk. Google scored a hit, many many years ago, but they haven't been able (or willing) to maintain that lead, and all their competitors have pretty much died out anyway.

If Google hadn't come along when it did, someone else would have stepped up. Maybe Altavista, or Yahoo, or someone else. There was a need, and a provider to address that need. The only reason we don't have a new search engine to beat Google today is because, well, everyone is scared shitless of going head-to-head with Google, except Microsoft with their propaganda-laced Bing embarrassment. They're just not the golden child people seem to think they are.

Re:hmm

[CharlyFoxtrot]

Altavista worked fine, HotBot too. I started using Google primarily because of the cached pages, not because the search was that much better. Plus like you say the Google interface was a breath of fresh air.

Re:hmm

[ToadProphet]

Everyone is in a good mood. Why not :-)

Modded Troll... now that's delicious.

Re:hmm

[mysidia]

Yahoo was originally a web directory, not a conventional search engine. The search results were provided by others.

In 2000, they signed an agreement with Google, and Yahoo's search was powered by Google, in other words -- if you used Yahoo, you were using Google.

That didn't change until 2005, and after several other search engine company acquisitions, when they developed their own search technology.

Re:hmm

[negRo_slim]

Plus like you say the Google interface was a breath of fresh air.

Sometimes I wonder if Yahoo hadn't made their default page http://search.yahoo.com/ early on, if they wouldn't have done somewhat better for themselves.

Re:hmm

[NekoYasha]

It's called a " running gag".

Re:hmm

[Afforess]

If you repeat a word too many times, quickly, your brain become tired of that word and it begins to become foreign to it. This event is similar in nature to looking at grid illusions. Your brain becomes tired after a few moments and you see dots.

Re:hmm

[jcnnghm]

You know from June 2000 to February 2004 Google was the backend for the Yahoo web page search. That was back when Yahoo was a web site "human directory" search first and foremost, and only secondarily a machine-powered internet search. Sort of like how Yahoo search is going to be powered by Bing in the future, and was powered by Inktomi before Google.

Re:hmm

[superdave80]

Funny
Funny
Funny
Fun...

Shit, you're right!

Re:hmm

[lawpoop]

If Google hadn't come along when it did, someone else would have stepped up.

Doesn't change the fact that it *was* them, who was able to do it when nobody else had been able to. So I think that yes, they did contribute a lot to open source development. It's not enough to have a good idea, or believe that someone will eventually get around to it; someone actually has to sit down and *do* it. If google hadn't done it then, we would be that much further behind in internet search technology.

Re:hmm

[Afforess]

Why is the OP funny, if it says "40% interesting," "30% funny" and "30% informative"? Shouldn't the post be "Interesting?"

Re:hmm

[Mostly a lurker]

Your recollections are different from mine. Prior to Google, I tended to use AltaVista and Hotbot. Searches took at least ten times as long. Results rarely included any recently created pages. The number of indexed pages was several orders of magnitude less than Google handles today (which in turn is one order of magnitude, or so, greater than current competitors). In spite of the fact that gaming of search engines is overwhelmingly targeted at Google, Google still does a relatively better job of finding the genuinely useful pages. Is Google perfect? No, of course not. Search is still only a partially solved problem. However, since its inception, Google has come up with most of the practical advances in the state of the art, as well as the best infrastructure for its implementation.

Re:hmm

[MobileTatsu-NJG]

They use the Linux platform to the absolute max, leveraging all the blood and sweat Linux developers poured into its development over the past 15 years, and yet, not contributing back any of their most significant enhancements.

Not contributing back!? Dude, they gave us *google*. Remember what it was like before google? When internet search was basically voo-doo crapshoots, that worked 25% of the time? They gave us a search engine that actually *worked*. Before that, you basically had to bookmark or memorize internet sites that you liked. Good luck actually finding what you were looking for without having an actual site in mind beforehand.

I think that alone has probably spurred the development of free software. Imagine being able to *find things* on the internet!

Are you kidding? Search for Quake? Porn. Search for a new version of Netscape? Porn. Google? PFtb. It always gave me Quake and Netscape. My pr0n searching was MUCH more productive before Google!

Re:hmm

[skine]

Woooooooooooooo!

(Apparently just entering "Woooooooooooooo!" creates an error. I have to explain that it's supposed to be a giddy mod, thus destroying any semblance of assuming intelligence present in at least part of the /. community).

Re:hmm

[Dahamma]

Because it's funny.

Re:hmm

[Anonymous Coward]

I CAN'T SEE ANY DOTS!

Re:hmm

[Dahamma]

You clearly weren't a daily Google user 10 years ago.

The moment I realized Google was completely superior to the others was when I was able to paste an obscure compile error for an equally obscure CPU architecture into Google and immediately get the answer back... the kind of utterly random error that a few years previous would have potentially taken hours to debug...

If Google hadn't come along when it did, someone else would have stepped up. Maybe Altavista, or Yahoo

And you were modded Insightful - sigh... So you are saying they decided "oh, well Google is pretty good at this - let's NOT STEP UP." Yeah, that's what companies do in that situation. Or maybe they do try, and fail (nothing wrong with trying and failing... but that's the REALITY of the situation).

Re:hmm

[Yvan256]

Move closer to your screen. There's plenty of them.

Re:hmm

[wdr1]

Put the crackpipe down!

I was an altavista user. A die-hard one, for most of the mid/late-nineties. In fact, I remember the day I finally convinced my boss to switch from Altavista to Google, because he had worked on Altavista.

Today's results completely blow away the search engines of 10 years ago. In fact, any of the major players -- Yahoo, Microsoft, even Ask & co. -- would blow away the search engines of 10 years ago.

(Add to the fact that the number of documents on the web that they need to crawl & rank have exploded.)

Your comment that "the resultant pile of URLs for any given keyword is utterly worthless" is itself hyperbolic nonsense. If that were true, nobody would use them.

Re:hmm

[HeronBlademaster]

I used Dogpile, back in the day; it would show you the results from ten or so other search engines.

Re:hmm

[unity]

The only thing you really missed there was the really simple, non-image intensive interface. That alone spurred people to use google.

Re:hmm

[PitaBred]

Only if it's ironing.

Re:hmm

[BikeHelmet]

Shit, I see dots from the start. My brain must be really lazy.

Re:hmm

[Anonymous Coward]

Many people think that Google's original claim to fame is PageRank. That's only partially true. Google became as successful as they are because of their systems-scalability work. That is, Google figured out how to build the biggest clusters, with the most storage space, the most computation capacity, and the lowest latency, for the least amount of money (compared to their competitors anyway). If you have 1000x times the computing power of your nearest competitor, then you can do 1000x as much data mining, which means that your search results (and ad relevancy) will be that much better.

For a long time, Google refused to release any information on their system infrastructure (it was their crown jewel, after all). The GFS paper was released in 2003, well after Google had put the filesystem (and its predecessors) to public use.

To sum it up: GFS has been one of the strongest contributing factors to Google's dominance. The idea that Google would voluntarily give this code to competitors is laughable.

Re:hmm

[sootman]

It really amuses me how all these different comments come up in every thread about search engines. Everyone's experience is different. Google is still very useful to me 99% of the time. As for AltaVista, I remember '96-'97 very well. I would usually use Yahoo first. If Yahoo only produced a small handful of results--literally, 10 or less, and no good ones--then I'd go to AltaVista and get tens of thousands of results. If I was lucky I'd find what I wanted in the first few pages, else I'd give up.

Google is still literally orders of magnitude than anything else I've tried. Disclaimer: I've pretty much used only Google for the last... um, however many years it's been since they came on the scene. I won't claim to have used it when they were still hosted at stanford.edu, but I heard about them early on (back when they had , probably from Slashdot, and I was impressed right away. I probably stopped using Yahoo altogether within a couple months.

Re:hmm

[V!NCENT]

It takes absolutely zero effort for this post to be modded funny

Re:hmm

[jimicus]

Of course, by posting under your user account, you're not a mod in this thread any more so it's all rather academic.

Re:hmm

[Vu1turEMaN]

Please take your meds, Rick Flair....

Re:hmm

[Xemu]

It takes absolutely zero effort for this post to be modded funny

It doesn't take much to be modded informative either.

Re:hmm

[MistrBlank]

Except for you apparently Mr. (Score:3, Informative)....

Re:hmm

[D Ninja]

C-C-C-C-COMBO BREAKER!

Re:hmm

[Mozk]

Almost every comment posted on this page is modded 5 Funny. o_0

Really, what's going on?

Re:hmm

[mR.bRiGhTsId3]

I've read the paper. You can definitely reproduce GFS given enough time from the paper, since all of the synchronization and master/slave node dynamics are described very well. The actual implementation seems like it would be monstrously complex. Not something 2 or 3 guys could pull off in a weekend by any stretch. Anyone who is actually curious, here is the original paper from the acm, though you may need a membership to view it.

Re:hmm

[mysidia]

Directory entries were submitted, reviewed, moderated, and maintained by humans.

Search Engines were automated, and very poor. At first they didn't have much indexed, then they got things indexed, but your search terms would turn up irrelevant hits, and you would sometimes have to dig to the 12th or 13th page to find the most relevant, most reputable / highest quality page.

A little later search engines got a little better, but started pushing sponsored results, and keyword spammers started flooding the first 10 pages of search results on popular search terms. The spammers of the world and "dot coms" became very good at manipulating search engines, the result was when you searched, you got tons of totally unrelated hits that had been kw-spammed or put there through other trickery.

And thankfully, now we have Google. Which performs some decent filtering of much of the search spam to help downgrade the low-reputation irrelevant sites in search results. And also does an excellent job at finding highly relevant pages on quality sites.

Oh yes, and the sponsored links are clearly labelled, and separated from the normal search results; so it's not "if you're our sponsor, your site quietly gets the number 1 search position".

By the way, the Yahoo directory still exists, so you can see somewhat how it was like, if you perform all your web searches only on 'Search: The directory' and don't choose the new 'Search the web' option

However, even that has been improved over the years.

It's not really GFS

[mysidia]

It's GoogleFS.

GFS refers to the Global File System, which is commonly used in Linux clustering environments.

By comparison, GoogleFS came second, is basically a no-name filesystem unknown to most of the IT world, because it's not available for use, hasn't been released as a product, compared to the well-established global filesystem.

It would certainly seem like the Global File system would have priority claim over the name GFS...

So let's stop calling Google's filesystem, which we'll probably never get to use GFS :)

Re:It's not really GFS

[mysidia]

That is a problem that may be getting corrected by the IANA TLA registry :)

Re:It's not really GFS

[corsec67]

Tell that to the NWA. Wrestling and a wildlife foundation should be even easier to tell apart, they both aren't as similar as two file systems.

Google is IT done right...

[Alien Being]

but God help us all if they ever do turn evil.

Re:Google is IT done right...

[PhrostyMcByte]

Google is what happens when developers and IT talk to each other correctly. Normally there is a brick wall separating the two, with IT guys being at the mercy of whatever the well-meaning but typically oblivious (to IT problems) devs cook up.

Re:Google is IT done right...

[ObitMan]

not on your life.
Developers constantly ruin perfectly good infrastructure.

Re:Google is IT done right...

[Nefarious Wheel]

Developers aren't IT?

Not really, no. It's kind of like the difference between a doctor and a patient. Or to use a car analogy, the difference between being an automotive engineer and the guy who takes money for candy bars, magazines and fuel.

Disclosure: I was a developer for about thirty years before I took a step down and moved into marketing. I learned a lot of languages but was stopped when I discovered I was having trouble mastering Hindi.

Re:Google is IT done right...

[Antique Geekmeister]

Because they're not allowed to share their ideas with IT, and vice versa. I can't list the number of times developers have published brain-scrambled vomit as part of their projects, because it didn't interest them and no one with experience was around to explain the inevitable problems. The maintenance model for subversion where you have to completely rebuild the repository to completely delete an accidentally stored DVD image is a classic example.

Conversely, I've expressed extreme doubts about projects that turned out to be effective and workable because my knowledge of file system behavior or hardware limitations was 3 weeks behind the times. I even spiked a project once for such reasons, although when the developer and I spoke without the confused manager in the way it became clear that the hardware _could_ support his needs.

Re:Google is IT done right...

[Duncan3]

Not really, it's IT done by not letting anyone over 30 or with any experience into the room. Every single issue they had to learn and fix mentioned in the article is quite literally standard textbook stuff in distributed systems, and has been for over 40 years. The failure model, the huge chunk sized, the single master problems... etc. Nobody who had taken even one decent class would have ever considered the original design viable.

They really should just stick to buying their tech pre-made like everything else Google is known for - acquisitions. Other companies are willing to hire experienced people. You know, those old lazy bastards that only work 40 hours a week because they have families, cost way too much to provide health insurance to, but get things done 5x as fast because they have done it before :)

Re:Google is IT done right...

[Daniel Phillips]

Not really, it's IT done by not letting anyone over 30 or with any experience into the room. Every single issue they had to learn and fix mentioned in the article is quite literally standard textbook stuff in distributed systems, and has been for over 40 years. The failure model, the huge chunk sized, the single master problems... etc. Nobody who had taken even one decent class would have ever considered the original design viable. They really should just stick to buying their tech pre-made like everything else Google is known for - acquisitions [wikipedia.org]. Other companies are willing to hire experienced people. You know, those old lazy bastards that only work 40 hours a week because they have families, cost way too much to provide health insurance to, but get things done 5x as fast because they have done it before :)

You hit the nail right on the head. The original GFS is pretty lame, as Google folks freely admit (full disclosure: I'm a fomer Googler, but I'm not telling you anything you can't find on ahem Google). The new GFS will also be pretty lame, because as you correctly point out, Larry, Sergey and Eric don't quite get the concept of experienced people who have done it before. All that standard clustering stuff has to be reinvented by Googlers, who frankly, have gotten a little soft over the years, now so used to working. We will see, but I'm predicting that the new GFS will still be a research project two years from now.

Curiously

[ShooterNeo]

In the article, it's stated that the load on the google file system has grown orders of magnitude greater than it was ever intended to handle. And one of the algorithm changes is that the chunks in the new file system are 1 megabyte in size rather than 64 megabytes. This is to reduce latency, which makes logical sense...but dividing a gigantic database into pieces that are 64 time smaller doesn't make intuitive sense...

Re:Curiously

[Alien Being]

"..but dividing a gigantic database into pieces that are 64 time smaller doesn't make intuitive sense..."

It does if it was 64x too big to begin with. Live and learn.

Re:Curiously

[Russianspi]

I don't know if that is so much the question, as much as: Would a highway get more people to where they are going faster if it were semis only or motorcycles only? The answer to this question (and probably to Google's question, too) is "It depends on a lot of other factors".

Re:Curiously

[martin-boundary]

It does if it was 64x too big to begin with. Live and learn.

No need to learn. 64x should be enough for anybody, dammit!

Re:Curiously

[Runaway1956]

The question is based on assumptions. I've personally pushed an 18-wheeler over 150 mph. I've pushed a bike over 170 mph. In both cases, the limiting factors were all the 4-wheelers. Take the cars off the roads, and let the bikes and the trucks run.

Oh yeah - one more thing. Mandate that cop cars have square wheels. They already have radio, they need a handicap to make things fair.

Quality of comments going downhill...

[s0litaire]

There's over 25 comments and not one has attempted to call it "Goatse File System"!

Whats up with you trolls! You guys on a union break or what!!

Re:Seriously Folks

[MikeBabcock]

Google has competitors?

Seriously, Microsoft has been promising a database driven filesystem for its server OS for years without delivering anything substantial to date, and it doesn't seem like they're running anything different internally either.

Where's meta-moderation when you need it?

[nobodyman]

I'm impressed that all of these Reddit users had the attention span to stay long enough to get mod points. But nobody likes a guest who overstays their welcome. Besides, I think somebody posted an animated gif of an old man falling down or something. GO CHECK IT OUT!!!1!1one

Re:GFS?

[Vectronic]

Wrong, its the Google GFS File System!

Better back your files up

[kerskine]

I have not doubt that the new file system will be great, but after reading this summary, the first thought I had was that I should back-up all that Gmail before they cut it over. I've been putting it off for far too long, but I'll just have to burn a couple of days of attention to do it.

Re:Better back your files up

[fracai]

Couple days?

Install OfflineIMAP or Getmail, write the config file, schedule cron, launchd, etc., and be done with it. Shouldn't take more than an hour. The bulk of that should be working through the config and both products include healthy examples.