Re: Search by urn ??

--- In [email protected], Arne Babenhauserheide <arne_bab@...> 
> As far as I know LimeWire uses Kademlia technology (DHT) for
Hash-searches, 
> which is more efficient for non-keyword searches. 
> 
> For searches by keyword they use Gnutella, for searches by hash they
use 
> Kademlia, a strategy which sounds quite useful to me: Gnutella is
heavily 
> optimized for searching by tags, and using it for hash-searches
would likely 
> reduce its efficiency for keyword based searches. 
> 
> Kademlia on the other hand is very efficient for hash based
searches, because 
> its whole network structure builds on hashes - each node has a hash
as ID, 
> and it maintains info about the files whoose hash is most similar to
its own 
> hash (similarity: xor). 
> 
> I think that's quite a nice way of keeping with the "one network"
philosophy: 
> One tool for one task, and Hash based searching is fundamentally
different 
> from keyword based searching. 

BearShare treats a hash much like a keyword and no special optimizing
was needed to accomodate it.  The reason LimeWire resisted was that
their QRT was only 64 kilobytes and they felt that including hashes as
keywords in the bitmap would saturate it too early. That may be so
because the tables are already flooding to 100% regularly without the
hashes. Even without putting hashes in there, it would be pretty
helpful to try to use larger tables now.

Still, if they have an alternate way of responding to hash searches
then perhaps they can let their ultrapeers respond with gnutella hits
to BearShare and other servents now that they have a way to find
hashes. It may seem to be overly generous to the rest of the network,
but none of the developers are overly competitive or anticompetitive,
so there's no reason not to try it now that all the tools are on hand
and in the software.