Re: Google Architecture

A slightly different spin. Perhaps the model for Google and the other
mega-search engines actually represents a niche market: the gleaning and
somewhat orderly presentation of random atoms of the world wide web.
That's a gargantuan 'niche' in topographic scope, but narrow in this
respect: a query only needs to get back an answer that satisfies the asker
within the first few dozen guesses. I'm always amazed when Google
satisfies me within the first handful of guesses, but I don't demand
instant gratification. I'll browse through a screen or two, then try to
reformulate the query. I can quickly skip over most off-base guesses, and
I tolerate repetitive guesses that show up along the way. But since each
query costs me nothing but time, I can afford to circle the wagons and
chip away at the chaos until some semblance of intelligence appears.
Furthermore, my motivation for asking is often recreational: scratching
the curiosity itch or settling a friendly dispute. I don't gamble, so
front line profit/loss is seldom a factor.

Meanwhile, Google makes money not from the answers they shovel my way but
from the advertizers who pay to peddle their wares on page displays. The
more times I reformulate an unsatisfied question, the more hits Google
racks up. This is not a conspiracy, just the way it works. I keep at it
because Google usually delivers. Eventually. I'm induced to persevere
because every query costs me the same: nothing.

Contrast this model to a traditional business enterprise I once worked at.
(Then) TRW Business Credit service collected accounts receivable data from
all manner of suppliers large and small. A potential creditor would query
the financial soundness a new customer by pulling a report of how well the
buyer was paying bills to other suppliers. Standard inquiry stuff, but the
wrinkle was the complexity of matching queries with account records in the
data base. ABC Pipe Fitters of Duluth may sound like simple look-up, but
subtle variations in name, address, and even supposedly unique entity
numbers made gathering up the 'guesses' a dicey proposition. At stake here
was not cocktail party one-upmanship but serious matters of money and
reputation. If a report listed erroneous matches, the fallout could be
severe: failure to note payment problems or successes could cause nasty
results either way. Same for failure to list true matches: news too rosy
or unfairly bad. On top of all this uncertainly, the report was paid for
by the requestor. Two reports cost twice as much as one. And two reports
on the same business had better be identical or the requestor would be
hopping unhappy. We employed several folks not then prominent in the IT
industry to exercise and hone the search mechanism. It was a serious foray
into fuzzy logic before I first read that term in print. And it all ran on
MVS.

Would the Google model have worked in that business environment? Not with
the flaws pointed out in this thread. How about querying your bank or
credit card online? Your utility usage? Your school records or the
maintenance history of your car? In such environments, what you demand is
a few good answers.

So, other than the 'niche' market of roaming the web in search of stems
and pieces of information ranging from titillating to useless, what real
business would put up with Google's imprecision? Who could afford to?





"Craddock, Chris" <Chris.Craddock@xxxxxx>
Sent by: IBM Mainframe Discussion List <IBM-MAIN@xxxxxxxxxxx>
06/04/2006 01:47 PM
Please respond to
IBM Mainframe Discussion List <IBM-MAIN@xxxxxxxxxxx>


To
IBM-MAIN@xxxxxxxxxxx
cc

Subject
Re: Google Architecture






<snip>

Of course, if you're indexing the entire internet those low-cost
thingies can still add up to big numbers, but it would be a mistake
(IMO) to assume we could to it any cheaper/more efficiently with a
traditional mainframe-based system design. It might be possible, but
there are reasons to think it might not go well for the traditionalists.

<snip>

CC



----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to listserv@xxxxxxxxxxx with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html

.

Relevant Pages