On Sun, Dec 31, 2000 at 03:48:04PM -0800, Ian Clarke wrote:
> On Sun, Dec 31, 2000 at 03:03:07PM -0800, Aaron Voisine wrote:
> > It definately makes more sense from a deniablity standpoint to
> > use a dropthrough encryption algorithm on the KSK first.
>
> We are smarter than we look ;-)
>
> > As far as being a broadcast, I understand that's bad, but I
> > figured with caching and low hops to live, it would be no worse
> > than any other data request.
>
> The primary means of ensuring scalability isn't really caching so-much as
> intelligent routing.
>
> Here is how I think searching should happen:
>
> We define a function that, given a list of key/data pairs (metadata
> describing the content) A, and given two other pairs, B and C, it can
> determine which of B and C is closer to A*. There are an infinite number
> of ways that this could be achieved, but the closer it is to what a human
> might do, the better. Someone did write some code which did this but I
> have no idea where they went.
>
> Once you have done this, you can treat the metadata for a piece of
> information as a key, and can route requests towards that key.
>
> The only remaining change is that, right now, a key will only be
> retrieved if you find it exactly, where as that is not going to happen in
> most cases with metadata searching. We need a new "Find Closest" message
> which will return a list of the closest keys to the one being sought
> encountered by the search. This would be returned to the user who would
> then select the one they want to try to retrieve.
>
> * In actual fact it will be more complex than this, since this comparison
> is likely to be a costly computation we will want to cache and
> precompute whatever we can
>
> Implementing searching will be great fun - if anyone wants to do
> interesting stuff with Freenet, I suggest that you attack this. To begin
> with, you can start to think of good ways to do a fuzzy match of key/data
> pairs to another set of key/data pairs. Of course, if you wanted to be
> super-smart, you could figure out a way to do a fuzzy match to an SQL
> statement (which rather than returning "yes" or "no", returned a measure
> of how close the match is).
The problem with searching is that it will probably be very difficult
to do it in a really secure manner. Figuring out how to do fuzzy
searching is easy - figuring out how to do it in a way which won't
provide plenty of both in and out of band information about what is on
which nodes is a completely different story. I essentially shot down
Aaron Voisine's keyword searching proposal (which even without this
wouldn't work) because it provides way too much information about what
is on which nodes.
--
Travis Bemann
Sendmail is still screwed up on my box.
My email address is really [EMAIL PROTECTED]
PGP signature
