Re: Email Web of Trust - Problem Statement

At 08:58 AM 3/12/2004, Yakov Shafranovich wrote:
>Mark Baugher wrote:
>
>>At 02:02 PM 3/10/2004, Alan DeKok wrote:
>>
>>>Mark Baugher <[email protected]> wrote:
>>> > So is the member of the web of trust like a maven, who passes 
>>> judgement on
>>> > the goodness of a domain based on its own idiosyncratic evaluation of 
>>> that
>>> > particular domain?
>>>
>>>   Yup.  We cannot realistically expect anything else.
>>
>>How about a MIB definition:  Let's say the peers who make up the "web of 
>>trust"
>>run the same protocol for authorizing incoming mail.  The protocol, whatever
>>it is, has counters for messages that are received but rejected as junk,
>>along with the address or name associated with the source.  The peers trust
>>each other to maintain these counters according to the MIB specification.
>
>I had been thinking along similar lines. What I had in mind is a web of 
>reputation system

I'm not sure we have bottomed out on what the relationship between "web of 
trust" and "reputation system" is so I'm not sure what a "web of reputation 
system" is.  I think of web of trust technology for MTAs as a means to 
extend a reputation system, which could be a single server or set of 
servers operated by a single service.  In the case of a web-of-trust 
reputation service, however, either single or multiple operators are 
trusted to offer reputation services.

http://www.stanford.edu/~sdkamvar/papers/eigentrust.pdf and 
http://lsirpeople.epfl.ch/despotovic/CIKM2001-trust.pdf talk about such 
systems in very general, peer-to-peer file-sharing applications.

>which is extensible via something like the MIB mechanism AND a base 
>MIB-like RFC defining neutral data points such as length of time MTA is 
>transmitting, average volume, etc. This way everyone can use the base 
>neutral data points and extend the web to include other things they might 
>be interested in.

I guess we can start defining it.  The antispam MIB is wide open and can be 
any number of things.  In the simplest case, it might be the number of 
times an MUA explicitly rejected a message from a bona fide source 
following user inspection.  There are all sorts of attacks to be considered 
including collusion among users, Sybils, etc.  And there is the problem of 
ensuring sufficient system integrity such that false positives don't smirch 
the reputation of mail senders.


>As a side note, I am not a big expert in SNMP but in theory can it be used 
>to provide information about MTAs?

http://www.faqs.org/rfcs/rfc2789.html

Mark


>Yakov