You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@spamassassin.apache.org by David Birnbaum <da...@pins.net> on 2006/12/12 16:39:52 UTC

BOT Army

Greetings,

I was reading the ideas about combating the distributed spam attacks, and I was 
wondering if some combination of a razor+distribution analysis of the IP 
addresses in the header would lead to a rapid identification of potentially 
infected machines.

If you think about the distribution of a normal email users, it's going to look 
like a very sparse matrix:

   (few IPs per sender domain) -> (few recipients per recipient domain)

A big email ISP might look more like this:

   (few IPs per sender domain) -> (many users per many recipient domains)

A spam army, though, is going to end up looking like this:

   (many IPs per sender domain) -> (many users at many recipient domains)

If one were to take the first couple of IP addresses in the header, and do DNS 
lookups of IP+recipient, the central DNS engines (not sure who or how those 
would get hosted) would rapidly be able to see a new machine enter the spam 
dominion and could start returning a value based on the distribution of DNS 
requests.

My graph theory is rather dusty these days, but if memory serves the 
connectedness and sparseness of the graphs is pretty low-overhead to calculate 
and maintain, so even at a relatively high load you could quickly start to know 
which IP addresses look "different" and thus return a higher score for them.

I don't know if anyone has pursued this type of approach thus far, but I thought 
I'd toss it out there and see if it stuck.

Cheers,

David.

Re: BOT Army

Posted by Duncan Hill <sa...@nacnud.force9.co.uk>.
On Tuesday 12 December 2006 15:39, David Birnbaum wrote:

> If you think about the distribution of a normal email users, it's going to
> look like a very sparse matrix:
>
>    (few IPs per sender domain) -> (few recipients per recipient domain)
>
> A big email ISP might look more like this:
>
>    (few IPs per sender domain) -> (many users per many recipient domains)
>
> A spam army, though, is going to end up looking like this:
>
>    (many IPs per sender domain) -> (many users at many recipient domains)

I believe Commtouch/DCC claim to do this.