NYT On Google's Role In Internet Advertising

prostoalex writes "John Markoff and G. Pascal Zachary from The New York Times take a look at Google, its already dominant position in the field of Web search and its increasing influence in the field of Internet advertising. Google is driving advertisers away from larger advertising venues, like AOL-TW et al., since (surprise!) people actually pay attention to relevant text links and are quite annoyed by pop-ups and similar "innovations". Some interesting data about Google: number of employees is about 800, number of buildings is 4, number of servers is 54K, for which there are about 100K microprocessors and 261K hard drives. This is claimed to be the largest computing system in the world, and that also raises barriers for anyone entering the field of Web search - most of companies out there can only imagine a Beowulf cluster of these, let alone build them so that the Web searches are delivered within a second."

Comments

The Link

[spoonist]

The REAL link to the article is this:

In Searching the Web, Google Finds Riches

NYTimes registration.

[termos]

It was starting to piss me off, so I created:

Login: sladotter
Password: slashdot

Feel free to use it.

Re:NYTimes registration.

[Odin's Raven]

It was starting to piss me off, so I created:
Login: sladotter
Password: slashdot
Feel free to use it.

I think this comes up every time a NY Times article is linked. Okay, my turn to remind people: If you don't want to register with their site, don't bother creating bogus accounts. It's a nice thought, but it's really not necessary.

Instead, just go to their archives section, where the articles are available without the need for an account. Just replace "www" with "archives" in the link. Example for this article:

http://archives.nytimes.com/2003/04/13/technology/ 13GOOG.html

TYPO IN ABOVE POST

[Uber Banker]

Opps, $10m is more like it, sorry typo, dunno how is did it twice... :\

But yes, $10m is not a lot for big corps.

Re:Google aren't big...

[packeteer]

Ok your are EXTREMELY far off. You may be a banker but you seem to not have experiance with technical situations. A company like google could end up spendind a million dollars on one person. Remember that to run an entire building is expensive. They have to pay all the people who many of which are expensive techs. Each server probably has thousands of dollars of equipment between the multiple procs and hig performance hard drives. The computers had to be setup of course which im sure costed a bundle as that would be no small task. Maintenance of the equipment would be a bitch. I doubt anyone could do what google has done with less than $100 million.

Re:What do you know,

[BrookHarty]

Hey, people do click those ads, they do respond to spam. Look at Iwon.com, they are still in business and they pay people money to use their search engine!

It is not a beowulf

[telemonster]

It is not a beowulf cluster, it is a distributed set of systems. In a beowulf cluster the memory is shared between hosts over "fast" networks connected to all of the peers. 54k is an awful lot of servers. How many SGI Origin 4000's running 512 CPUs per cabinet with it's high bandwidth I/O subsystems (disks and networking!) would it take to do the job of Google's cluster? Would there be benefits to managing 20 1048 CPU single-OS systems versus 54,000 linux machines? Other than the obvious fact that Linux tends to get you lots of press where as conventional well engineered unix systems don't? Archive.org also uses a similar distributed model, adding servers as their archive grows.

Re:It is not a beowulf

[dubl-u]

Would there be benefits to managing 20 1048 CPU single-OS systems versus 54,000 linux machines?

Some, but they don't outweigh the costs.

A 2 GB 2 CPU box is well under $1000, right? so 32 of those would be, say, $25,000? But the SGI Altix 3700 configured with just 64 processors and 64 GB of RAM costs a cool million dollars.

Even if the multi-box solution demands more fuss to manage (higher failure rates, I'd guess), since you've saved $975,000, you can afford a little admin time. Plus, the cost of somebody who can build intel boxes must be less than half of somebody who is blessed to work on high-end SGI supercomputers. And I'm sure the SGI service and support contract in't cheap, either.

And the nice thing about a distributed approach like this is that if any given box goes down, so what? Whereas if one of your bazillion-dollar 1048-CPU boxes coughs up a hairball, that's a noticeable percentage of your regional computing capacity.

There are some applications where you can't go for this federated approach, but not as many as people think.

Re:concentration of power worrisome? censorship?

[n__0]

I think recently this happened with an article on the register, where they coined the phrase googlewashing. Almost all the sites that subsequently used the term linked to the registers article but its place on googles rankings fell very quickly defying logic. The register mentions it at: http://www.theregister.co.uk/content/6/30195.html

Re:concentration of power worrisome? censorship?

[More Trouble]

This is all the more vicious since the user is not warned that certain sites are censored.

Nonsense. Search Google for "scientology+leaflet". Scroll to the bottom of the page. Note the warning. Note that the warning links to the list of removed links.

Concentration of power is worrisome. But complaints should follow a problem, not a concern.

:w

The barriers to entry...

[cperciva]

... are not quite as high one would think. Competing in the search engine business does not require having the largest network on the planet.

For that matter, you don't even need to have the largest index. Not all sites are created equal -- an index of even just 10% of the web would satisfy most people, as long as it is the right 10% and is searched effectively. There are some advantages to being complete (non in googlis est, ego non est) but for common place searches it isn't necessary. While indexing the entire web may be very expensive, indexing a small *useful* part of the web is much less so.

The largest barrier to entry is simply the problem of coming up with a better way to search. Google has a very effecting algorithm, and they've got lots of smart people.

Re:The barriers to entry...

[goon america]

It's really the wrong term. It's not barriers to entry, it is economies of scale.

In other words, as the company gets bigger, the cost per customer decreases relatively -- increasing returns to scale. So, larger firms will always have that advantage over smaller ones, which makes it difficult for a new company to enter the market successfully.

There are different conceivable curves of (num. customers * cost/customer), so it may become flat or go up at some point, meaning that there will probably be several competitors which must be of a certain size. If the curve continuously approaches 0 cost/customer, then game theory would lead us to believe that only one firm would be able to compete in that market.

Re:Click-Through Ad Pricing

[wantedman]

Besides being rathar Shady, I doubt your trick may work.

The advertiser can set a limit where Google stops placing their advertisements, hence stops any click thrus...

Tracking IPs is pretty much standard for Click Thrus pricing.

Billing is done thru CPM, which means amount of views / 1000, so as long as you don't exactly match the stopping point, the advertisement will get more views, lowering the cost of the ad.

Adword Pricing info

Re:Well there's just one thing missing right now .

[mindstrm]

The linked article states they filter results based on location for LEGAL reasons. What do you want them to do? Lose the ability to do business in other countries?

The mere fact that they COULD adjust results for some other reason, based on location, just speaks for the sophistication of their system.

Alcohol ruins a lot more lives than pornography does.

Re:another (unsubstantiated) google fact!

[ePhil_One]

The electricity consumption on a modern PC is about 60-90 watts or about $25/yr for electrity. It would at least pay to have a low end staffer or an intern to go digging them out. Especially when you consider the cost in rent for a thousand dead machines. 20/rack and each rack is about a yard sq. Rent is probably a few dollars per sq foot per month. The costs just start piliing up when you factor in cooling and all of the other ancililary factors.

Chock full of bad assumptions.

1) The CPUS are dual CPU/big RAM/Dual drive, so a better budget is likely 150-200 watts. But...

2) The machines are not local. So you have to pay a tech to visit the machines. Airfare, hotels, etc.

3) They are using VERY dense machines, they get more like 60 systems (120 CPUs) to a rack.

4) The are using hosting facilities, such as Exodus/Abovenet/other survivers; these guys are chargeing a flat rate for Bandwidth, power, space, cooling, etc.

So the net result is you gain almost nothing by powering down a bad system, except the ability to add a working system. If the cluster is not overloaded, there's not much ROI on having that extra power. So its just a math problem.