Slashdot Log In
Craig Silverstein answers your Google questions
from the my-server-farm-is-bigger-than-yours dept.
1) I've wondered
by lblack
Google always seem to be early-to-market with some really highly developed software solutions, and also always seems to have the backbone to support them.
I'm curious -- what drives the innovation? Is it the hardware team advancing architecture to permit the software team more room to play, or is it the software team saying, "Hey, look what we got!" and the hardware team dropping the iron to implement it?
I understand there must be some level of synergy, but is it completely seamless or is one side of the equation effectively driving the other?
Craig:
Actually, the innovation is driven neither by hardware or software, but by products. We look around and say, "What would be the next great product to have?" and then figure out what software and hardware we need to make that product work and work great. If that figuring goes along the lines of, "Oh, it shouldn't take more than two weeks to get the code ready for public use, so that should give us plenty of time to get the 2000 new machines we'll need ordered, delivered, and installed" -- well, that's the kind of environment in which innovation flourishes at Google. [:-)]
2) Network Management Tools/Technologies
by kaladorn
What technologies help to support the Google server farm? What kind of automated monitoring and trouble reporting tools are in use? Are they home brew, open-source, or COTS with some customization (scripts, etc)? And if you had to point to one area of network management and say "we could use some improvement or some better tools", what would that area be?
Craig:
Almost all the technology we use to support our server farm is home-grown. The system we've built is so efficient we can maintain more than 10,000 computers with a handful of ops folks.
Of course, we benefit a lot from our massive redundancy: Unlike many companies, we don't need to worry immediately if a computer, or two, or a hundred, die, because the dead computers have lots of clones.
The biggest issue when you have more than 10,000 computers is that network management tools based on visualization become inadequate to the task: even if the UI is very good, there's often too much going on (ie,going wrong) to work effectively. At this level, you really benefit from tools that can not only identify problems but fix them. Of course, it's hard to write general tools for this, since "fixing problems" is typically pretty application-specific.
3) As a market leader...
by Marx_Mrvelous
It's well known that you use Linux in your mega clusters. I was wondering if you have ever been approached by Microsoft, Sun, or HP in an effort to switch to their proprietary OSes.
I can't imagine that you haven't. It must have been a huge decision to invest in one technology, so are you satisfied with what you have?
Craig:
We have been approached by several vendors. However, the advantages of Linux for us are pretty strong: It's an environment our developers tend to be familiar with, it offers unsurpassed tech support (we usually talk directly to the author of a piece of code when we're having problems with it), and it's cheap -- an important consideration when you have over 10,000 computers.
I think Linux works here as well as it does because of our technology culture. Our engineers feel comfortable being a partner in debugging kernel problems. For companies that would like to be able to give bug reports like, "Our network is slow" and have someone else take things over from there, Linux probably is not yet the ideal choice.
There's also a question of "Why Linux rather than FreeBSD?" or another free unix-like OS. We're not really religious about this issue. We used Linux -- as well as other, proprietary Unix variants -- when still at Stanford and were happy with it. My guess is if we had used a different open-source, unix-like operating system, we would have been happy with that as well. We're pretty pragmatic about using what works well for us.
4) Google's inescapable coolness.
by rob_from_ca
How do you avoid business pressures to make short-sighted solutions, and consistently make good, common sense ideas work instead of adopting ones from marketing sources? Not only does Google have the best search engine technology, but you consistently do the "right" thing. Clean, quick homepage, text only well-identified ads, interesting research projects, etc...This is the way many search engines start, but they all went the way of the "dark" side instead of adopting the "right" solution. In my jobs, it's been very difficult to execute and justify good engineering (or just common sense) under pressure from the people who control the money. Any advice for driving through well-thought-out decisions instead of adopting the "management fad of the month"?
Craig:
You know, it's this kind of cruel, hard-hitting question that gives the press a bad name. But, rob_from_ca, I know you're not really a member of the press corps -- are you? -- so I'll let it slide.
I think you're right that it's easy for a company to start with a laser-like focus on user experience, but hard to keep it up as the company grows.
I think there are two important factors that have helped Google keep its focus on users. One is that the founders have stayed actively involved in the company. The basics of our company flows directly from them. Larry Page's background is in user interfaces, and that really shows in the design of the site and in every project we do. And both Larry and Sergey Brin firmly believe that if we concentrate on users, everything else -- including money -- will follow.
The other important decision, which I can't stress enough, has been hiring. We've hired people who not only agree with this user-centric view of the world, but embrace it. Knowing what I know now, I'm infinitely impressed by how much our VP of Worldwide Sales and Field Operations, Omid Kordestani, embraced Google's policy of eschewing banner ads in favor of text-based ads, using an advertising system we developed ourselves. It's paid off, but three years ago it was far from a sure thing.
5) Google and IP address.
by Anonymous Coward
Why in this day and age does google continue to penalize sites that are virtual hosted? With ip addresses becoming harder to get/justify every day why does google discount the relevance of links that don't come from a unique ip address. Please don't just deny it, I think the Internet community deserves an explanation.
Craig:
I can't just deny it? What are my other choices? [:)] Actually, Google handles virtually hosted domains and their links just the same as domains on unique IP addresses. If your ISP does virtual hosting correctly, you'll never see a difference between the two cases. We do see a small percentage of ISPs every month that misconfigure their virtual hosting, which might account for this persistent misperception--thanks for giving me the chance to dispel a myth!
6) Weighting of heuristics
by jolshefsky
As the web develops, methods of matching a set of search keywords to a set of websites related to those keywords must change with it. I envision that the Google algorithms rank search hits by summing weighted factors such as overall site popularity, META tag keywords, META tag descriptions, TITLE tag contents, text contents, keywords containted in URLs, and so on.
Can you talk a bit about how those weights have changed over time? Have there been any surprising shifts?
Craig:
a) I'm afraid not, and b) No comment.
7) Regression
by Have Blue
The Internet is always described as a distributed system with no single point of failure. Google, however, has quickly become by far the most popular method of locating information. "Surfing" has been killed with modern search technology, it's so much easier to look through Google than the Web itself. If Google was down, I'm sure the Internet would be far less useful.
Do you think Google has become an Internet point of failure? With the competition for larger and larger indexes, is the Internet becoming centralized? Do you think this is a bad thing?
Craig:
It's true the Internet is distributed, but Internet services have never been. We saw that really vividly a few years ago when Network Solutions had a screwup with their root nameservers. As I recall, the Internet was basically unusable until DNS got fixed up again.
I think the growth of search engines is a sign that, in fact, the internet (well, the web in this case), is not becoming more centralized. If it were, then people could use a centralized registry to find whatever they needed to know. As it is, information is spread out throughout the web, so only an index like Google can tie it all together.
8) Favoring Big Guys
by PenguinRadio
Does google's policy of "ranking" the sites that have hits favor the "big guys" over more specific smaller traffic websites? That is, would a story on a site like CNN get a higher ranking in google on a keyword "Gulf War" than say a site (gulfwarveterans.com) that deals 100% with the Gulf War? Do you think you are leading to the commercialization of the web (i.e. the big power players) over smaller sites?
Craig:
Hmm, everything I wanted to say here has already been said in the Slashdot discussion on this question.
But in my own words: Google doesn't actually use traffic ("hit") analysis in its rankings: the rankings are based entirely on how sites link to each other. One consequence of this approach is that sites like gulfwarveterans.com, which maintain a consistent focus on one issue, are more likely to accrue lots of links than a transient news story, even one on a major site.
Indeed, searching for "gulf war" on google turns up two Gulf War veterans sites in the top 5, including gulfwarvets.com.
9) Dot com changes?
by Telastyn
Last I heard Google was still the stereotypical "startup" type company; promoting morale over bureaucracy as long as the work got done. Hockey, pool, the Greatful Dead's ex-chef (iirc?), and tons of other perks.
Did google keep the atmosphere as you've grown? did they keep it while others tanked?
Craig:
We still pay a lot of attention to making Google a place people like spending their time. The latest is a massaging chair we imported from Japan, so people could get massages even when it's not our masseuses' regular working hours (and they use the chair, too!).
We set this up from the beginning. (Healthy Choice granola bars in the breakroom: that was Sergey. All the M&M's you could eat: that was me.) We still see advantages to it. We think these efforts help productivity rather than hurt it. When you and a co-worker discuss an idea in a conference room, that pretty much limits the communication potential to just you two; when you discuss it over a game of pool, soon half the company has wandered by and had the opportunity to comment.
10) Google's first programming contest
by PK_ERTW
Google recently ran it's "first annual programming contest," with a winner receiving $10,000. Many slashdotters suspect this was simply a way to recruit new talent. So, was finding new people one of the initial goals for this project, and have you hired any new programmers as a direct result of it? What were the other goals (PR, generation of new ideas, etc) where there?
Craig:
The main goal was to have fun and to get people thinking about what they can do with large quantities of information. If we got people excited about the field of search or data mining -- even if they never submitted a program to us -- then that entire area of research benefits, and ultimately Google benefits as well.
The fact that our terms and conditions mentioned that we retained unexclusive rights to whatever people submitted, hints at our attitude. If really good ideas came out of the program, we wanted to be able to use them. On the other hand, we weren't using the contest as a substitute for consulting or anything (or else we would have demanded exclusive rights). And if the authors of the good programs wanted to come work for us, so much the better. For people who were excited by this project, we already knew there was a cultural fit.
[The following question was added to "the list" by Craig -- ED]
11) Forget Craig
by Talisman
No offense to Mr. Silverstein, but I'm much more interested in Cindy [McCaffrey]! Beautiful, highly successful nerds are terribly rare!
Just so I'm not off-topic: Mr. Silverstein, how does Cindy look in tight sweaters?
Craig:
If you did any research at all, you know that Cindy is our Vice President of Corporate Communications. As such, she takes an active role in Google interviews, such as this one. In fact, she's looking over my shoulder even as I type this. And, let me just say, she ... Hey Cindy, what are you doing? No, don't press that button! Hey! erwqu8poxasewrvNO CARRIER
Web services? (Score:2, Insightful)
"Yellow Pages" for the Web services effort? What about one uddi.google.com?
Good luck!
public access to raw archive (Score:1)
Good interview... (Score:3, Interesting)
...but I really really wanted to know what their electricity bill was!
I hadn't thought about the inherent redundancy in Google's design before, so if one or a dozen boxes go down the system as a whole remains virtually intact. It's an interesting effect of that level of decentralization, I guess. Maybe they should be a model to other organizations with an eye towards security and survivability?Re:Good interview... (Score:4, Interesting)
If they are co-located in a server farm, power is normally included with the fee per machine.
In either case, there might well be no power bill to pay!
One interesting unasked question: Could they distribute their servers geographically, through several server farms in widely distributed points, to increase redundancy in case Al Queda decided to target their server farm? Might be a thought.
D
Question for the google guy... (Score:4, Funny)
A: No.
Re:Question for the google guy... (Score:5, Interesting)
I had no idea of the scale (Score:5, Interesting)
I knew Google was into clusters, but holy crap: over 10000 computers?!? Do other clusters of this size exist elsewhere?
This makes the lack-of-ads-down-your-throat aspect of Google all the more bewildering. Their electric bill alone, must be enormous.
Obligatory (Score:5, Funny)
Low brow trash (Score:2, Insightful)
Please masturbate before submitting questions, not during.
Come on guys, clean up the crap and treat women with respect and you might just get to have sex with another human being for once.
Re:She is not a babe (Score:4, Insightful)
Give me a break...there is nothing at all wrong with this lady.
Interview loophole? (Score:2, Interesting)
Is there a way to prevent this?
What happens when trolls wait and upmod something you don't want to ask the interviewer?
Shouldn't the moderations stop for interviews after it leaves the frontpage? Or was this an editor moderation??
No disclosure about ranking methodology... (Score:1)
How about telling us how to do virtual hosting? (Score:5, Interesting)
Question #6 (Score:5, Informative)
http://citeseer.nj.nec.com/
very good interview as mp3 (Score:2, Informative)
dialup connection (Score:5, Funny)
Like he would work over dialup from one of the highest bandwidth offices in the world?!? They must have been at a hotel together. :-)
No Network Stats?? (Score:3, Interesting)
The Pigeons (Score:4, Funny)
Now, I am stuck with the thousand monkeys we bought clamoring on the keyboards churning out shakespearean sonnets.
Re:The Pigeons (Score:4, Funny)
Is your network of monkeys RFC-2795-compliant? [ietf.org]
Google a database. (Score:1)
Cindy (and some other people...) (Score:1)
We played pool at Eudora, but the beer was better (Score:5, Insightful)
We had a pool table at Eudora, when we were in Building I (the one by the cemetary, ironically enough). It got used a lot, although there was very little work talk that went on around it. Some, but not much. Mostly it was guys from tech support on break. More discussion took place when they had beer busts out on the lawn, or when they brought in food late at night for all us guys lucky enough to be working late during pre-GM release crunches. In fact, they wound up taking the pool table away because it was causing "productivity" to slip.
The beer busts worked particluarly well (and not just because I a) like beer or b) would wait until they were over and then would stuff the leftovers into the fridge in my office). It was like 3 hours of sitting around in the grass, engineers talking to sales, marketing talking to project mgmt, everyone talking to everyone else, VPs and admin clerks mingling. We couldn't all get together without talking about work stuff at a bar during happy hour, much less on a Friday at 2 in the afternoon on the front lawn of our building. The freebie dinners wer egood, but only engineers got to eat, so the group was more limited. We didn't get the cross-polination that the beer busts had.
If had to recommend a "dotcom-ish" group activity, I'd say a nice summer afternoon, some grass, a few frisbees, t-shirts for the employees-of-the-month, and a big tub of beer, wine, and soda. Very informal, just come and hang out. That's a really good way for a department head to get feedback, and way better than "all hands" meetings. I remember one day we had an all-hands, and not one person asked a question. A couple days later at a beer bust you couldn't get a word in edge-wise with the VP he was so busy talking with people.
Anyway, gone are the salad days...
-B
Not that kind of grass (Score:5, Funny)
Uh, that came out kinda wrong. I meant the kind of grass you sit on and the kind of frisbees you throw. I'm surprised I didn't mention patchouli and a hackysack. Jeez. At any rate, I wasn't advocating getting the whole company stoned. Although it probably would have helped in our case...
-B
Dejanews (Score:1)
download photo of Cindy (Score:1, Insightful)
Google Image Search on Cindy (Score:5, Informative)
JOhn
Unprofessional! (Score:1, Insightful)
Craig:
a) I'm afraid not, and b) No comment.
Wow, that is pretty low Craig. Couldn't be bother to even touch on the subject of how google works, huh? You couldn't be bothered to share some surface details about the weighting methods. About why method have changed. About why certain method work at one point in time but not another.
You just stiffed the entire
This is not a troll, but that responce was the most pricky I have ever seen in a
Bandwidth (Score:3, Interesting)
How many queries do they server per minute/sec ?
That was refreshing (Score:1)
With all the news about immoral corporations sometimes it's easy to forget that there are good ones out there. And it's always nice to be reminded that doing things right does pay off. Keep up the good work.
i still want to see it. (Score:2)
just one picture
I can vouch for that! (Score:2)
by PenguinRadio
Does google's policy of "ranking" the sites that have hits favor the "big guys" over more specific smaller traffic websites?
I can certainly vouch for Craig's response. My own site comes up as #1, and it's only three letters
I think I have to hate him now (Score:5, Funny)
Sure, and while you're at it why don't you tell us about your Ferrarri and the supermodel you're banging. Bastard.
Electricity Bill (Score:1)
If they each had a 14 inch monitor that was always on, it would double the bill annually. I can't imagine they would have one monitor for each machine, that would make no sense. The way I figure it, they still have one heck of a bill each month when speaking from a Joe Schmoe standpoint, but I don't think that they would have any trouble meeting that bill with the traffic they get and their limited advertising. Of course that bill is just for the computers themselves, not to mention the cooling infrastructure, lighting, etc... Anyone have any experience in the area of cluster cooling and the expenses associated with it? I know I saw something on Discovery or TLC about the units used in the old days, but what about now?
All of my statistics came from here [pcworld.com].
Hmmmm (Score:1)
how does Cindy look in tight sweaters (Score:3, Informative)
Cindy [McCaffrey]... (Score:3, Funny)
couple of things I didn't see (Score:3)
I'm also interested in seeing a discussion with actual google people on their policy on what's "suitable" and "unsuitable" as far as advertising goes. I've heard that pro-gun sites are considered "unsuitable" for no other reason than someone at google doesn't like guns. Google's right, but one starts to wonder if those biases are going to get built into the search engine algorithms sooner or later.
As google becomes a more and more important tool for getting to the rest of the Net, their politics are an issue whether they like it or not.
He didn't answer the question posed! (Score:2, Insightful)
Let me put it an other way. Consider that some terrorist manages to destroy all the Google servers. What happens to practical day to day work that uses the Internet?
Craig, why did you duck that one? (Score:1, Redundant)
That didn't actually get answered.
I know Google's been somewhat Akamai'ed, which helps, but if for example your data warehouse had a 'plane land on it (this kind of thing has happened a lot elsewhere in recent days), how long would Google be off-air for?
NO CARRIER (Score:2)
cruel? (Score:2)
Oldskool (Score:2)
Eh, sonny? What's this DNS you're talking about? Only sissies use domain-names, real die-hards use IP-adresses. Now, lets see, what was that address for google again
Linux Support (Score:2)
Why is the standard American workplace so stale? (Score:1)
Why do most American executives think so differently? When will most industries realize that not just customer support/loyalty is important, but employee morale and out-of-the-box thinking leads to more productivity (when done right) and then more profits (if managed right)?
Yes, I'm jealous.
Re:Grrrr (Score:3, Informative)
Re:Grrrr (Score:2)
http://images.google.com/images?q=%22Cindy+McCaff
Re:Grrrr (Score:4, Funny)
Duh... use google [google.com] :)
You give an obvious answer, but you don't ask the obvious question: What is the significance of the "I'm Feeling Lucky" graphic included in that collection of images?
Cindy googlewhack! (Score:5, Funny)
>
> http://images.google.com/images?q=%22Cindy+McCaff
Sure, that's a good start...
On the bright side, "Cindy McCaffrey" sweater" [google.com] is a Googlewhack, albeit not a pure one.
This brings two obvious followup questions:
To Craig: when will someone in your office leak us a copy of the .GIFs showing Cindy's meteoric rise in Google's images zeitgeist this week? Can we see it in the next Slashback, please?
Also to Craig: in how many hours, and how hard, do you think Cindy's gonna kick your ass for handing her over to the raving /. hordes as the older, classier, and way more-reasonably-hairstyled, replacement for our mad collective crush on Natalie Portman?
Finally, to Cindy if she's crazy enough to admit it: Do you like grits? ;-)
Re:Grrrr (Score:1)
A link with a picture (Score:2)
No nekkid pictures though - I keeping them to myself
Re:How did that lame 'tight sweater' question... (Score:1, Flamebait)
Sigh; It _was_ a good joke until you wrecked it :) (Score:1)
Seriously, I bet they had some fun tracking how much of a hit response they got on their exec profile page in response to that post. Got me at least, she's kinda a hottie
G
Re:Grrrr (Score:1)
I thought question 6 was pretty funny, too.
Re:How did that lame 'tight sweater' question... (Score:1)
Sorry.. did you read the interview? It was plainly stated that Silverstein (a) reads Slashdot (and therefore could have found said question himself) and (b) chose the questions himself.
To rephrase it was not submitted to him, he chose it.
Again: not submitted, chosen
Wow, I wonder sometimes if the average Slashdot reader can write but not read.
Re:How did that lame 'tight sweater' question... (Score:2, Informative)