Bob La Quey wrote:
On 9/14/07, Andrew Lentvorski <[EMAIL PROTECTED]> wrote:
Bob La Quey wrote:
Well it does raise an intriquing question. Where do you
distribute search? I see no reason for search to be centralized.
The net was originally intended to be peer to peer. I see no
real reason that it is not improved by returning to those
roots.
I do. There is no good method other than "flood" to search p2p. (Who
has it? Dunno. Who is closer? Dunno.)
Perhaps a heirarchy would help? A bit like DNS. Your search term
goes into a heirarchical system and is only sent to those boxen
that have the relevant index.
Okay, so that's an index to an index. So, how do you maintain the
accuracy of the index to the index?
It's turtles all the way down ...
Kademlia attempts to answer this. However, again, the problem is that
Kadmelia assumes altruism, equality and accuracy in the nodes. These
properties are very absent in a system in which there is high incentive
to game.
And we haven't even gotten into malicious nodes or nodes that choose not
to cooperate.
The problem is that the only person with incentive to have you search
accurately is the content provider. To everybody else, you are a drain
on resources (at best) or someone to be messed with (at worst).
Maybe you have not looked in the right place. Damned if
I know but I have known some extremely intelligent people
who spent huge amounts of time looking in the wrong places.
Well, that is often a good place to find a new answer. ;)
However, I'm not sanguine about much improvement.
P2P vs. centralized have different strengths in different domains. When
searching for something *specific* we tend to use a centralized index
(libraries developed card catalogs for a *reason*, after all). When
searching for something vague, we tend to use P2P (You ask another
person, "I liked book X. What other book might I like?")
This seems very similar to the Google vs. de.too.many.periods.us dichotomy.
-a
--
[email protected]
http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list
