On Sun, Jun 03, 2001 at 01:25:14PM +0200, Stefan Reich wrote:
> I see a problem with distributing the search queries and answers through the
> network.
>
> In Freenet, requests for keys are not being broadcasted, but sent to a chain
> of nodes
You don't say ;-)
> If one of those nodes knows the key, the search
> is successful. The direction the chain takes can also be influenced by
> individual nodes if they know in which part of the network a specific range
> of keys might be found. This is what makes Freenet scale.
And the search mechanism I described takes advantage of the same
mechanism.
> How does this translate to fuzzy searching? Suppose we distribute the meta
> information along with the data itself. Then, basically, the chance of
> finding the metadata you're looking for is just as high as finding regular
> files.
Not sure what you mean by this.
> But, for scalable metadata searches, we need nodes that specialize in
> certain types of queries (automatically, just like specialization in Freenet
> works for keys). This is probably feasible, but at least it requires a new
> architecture; the mechanisms that Freenet currently offers don't suffice. So
> I would say, separate fuzzy search from the Freenet core.
Please read my proposal carefully, it works in pretty-much the same way
as ordinary searching in Freenet.
> Basically, Gnutella is trying to solve the same problem we're discussing
> right now (except for the insert mechanism that doesn't exist in Gnutella).
> So we could learn from their experiences.
There experience has been a network which doesn't scale and lacks any
anonymity.
> Another thought comes to my mind (should have been a separate mail I
> guess) -- it is not good if data and metadata are usually on the same node.
> This will practically eliminate deniability. With metadata, you know what's
> in your data store; without it, you don't.
Did you even read my proposal? I discuss this.
> But still, the situation is different when the metadata is delivered to your
> node along with the data itself. Then you don't have to query any external
> indices. You could even write (or be forced to write...) a simple piece of
> code that scans the node for, say, *.mp3 and removes the corresponding
> files.
This is why I suggested in my proposal that metadata and data be treated
separately, perhaps even in separate datastores.
Ian.
PGP signature
