Re: Drop Dynamic Querying - It only benefits Spammers these days

From: "Arne Babenhauserheide" <[email protected]>
> > Running a trustable reputation system can become very costly, and this is
> > where the whole business of antispam-system providers are working to help
> > their customers. But is there someone ready to pay for this service, or
> > readyto support this huge charge for free?
> 
> Definitely not (at least not for us...), so we need a decentral solution, I 
> think. 

That's the problem: any decentral system will have spammers "collaborating" to it, and it can then be abused easily, and probably automated to create fake bad reputation for perfectly valid non spamming results; the net effect is that they can use it to give even more reputation to their own spam results.

The only way to solve this problem is to build a network where users do build their own network of trustness for some well-behaved agents that are collecting the live addresses of spammers, and to have servent vendors participate also to this system.

Such trust network needs secure authentication (or signatures in the data they transmit) to avoid trusted sources to be faked by spammers. That's where the huge cost is (think about server resources, and huge amounts of bandwidth used for data updates or for queries from servents everywhere on the Internet), and in the long term, it is unavoidable, like it is for fighting spammers in emails.

We can limit the impact of the cost by allowing several wellknown sources of trustness to collaborate, but even in this case, the system cannot provide complete proof that someone is not a spammer, or to avoid that some bad address is no longer associated to a spammer but to a legitimate source.

Trustness sources cannot be anonymous and must be verifiable using reasonnable reliable sources of information (for example DNS records, e.g. the inverse-resolution of their source IP address that provides their trusted domain name, or e.g. a forward resolution of a domain name given in the data signature, if the IP source is accessible, the latter allowing trusted sources with dynamic IP addresses but being less reliable). Og course we can reduce the number (and delay) of verifying the source if we have a cache of their secure signature certificate, and the signature scheme can itself be authenticated by another reliable source of trustness (like Certificate Authorities).

For lots of reasons, however, notably privacy but also for efficiency and higher security, the network for verifying trustness should be completely decoupled from the topology of the Gnutella network. This means a completely separated, out-of-band and desynchronized, protocol. This also means that the Gnutella protocol itself should not be altered to implement it. This will also allow choosing various protocols independantly.

I don't think that the Gnutella protocol should even attempt to transmit such information about trusted sources and spam sources: the servents will operate the protocol in parallel task/service, using a separate port, without signaling to others on the Gnet what they are silently filtering out (but of course, the servents may signal to their local user what they are filtering, if they wish, or could provide them with summary statistics...)