Re: Re: My Gnutella spam reduction idea revised

Am Montag, 7 de Mai de 2007 20:25 schrieb Michael Rogers:
> just_courtney_girl wrote:
> > Now, one possible problem I'd see with such an
> > approach is what if the spammers/robots
> > participate more than the users and good files
> > are marked as garbage and bad ones are marked
> > as good?
>
> I don't think Credence is based on voting - if I recall correctly it
> looks for users who have tended to agree with your ratings in the past,
> and asks them about files you haven't yet rated.

It also looks for people who have disagreed with your ratings and uses their 
votes inverted. So spammers need to vote up many files to be able to vote up 
one of their bad files and the good votes of many spammers in fact help the 
users identifying good files (because those will have been voted up more 
often than others, while the spam files will have been voted down by honest 
users quite often). 

The obnly thing which might twart Credence is spoofed sha1sums (so files are 
served which don#t fit the sha1), but that problem could be solved by relying 
upon TigerTree instead of Sha1, because to my understanding a spoofed 
TigerTree hash would be recognized when the first packet is downloaded (and 
doesn't verify). 

More information on Credence (and the conclusions of its authors): 
- http://www.cs.cornell.edu/People/egs/credence/Credence_nsdi.pdf

More papers: 
- http://credence-p2p.org/paper.html

Sourcecode: 
- 
http://credence.cvs.sourceforge.net/credence/LimeWire/core/com/limegroup/gnutella/credence/
- 
http://credence.cvs.sourceforge.net/credence/LimeWire/gui/com/limegroup/gnutella/gui/credence/

Best wishes, 
Arne


[Non-text portions of this message have been removed]