[milters] Archive

Lists Index Date Thread Search

Article: 265
From: Anthony Howe
Date: 2004-11-29 10:56:18 -0500
Subject: Re: milter-mole

Removal...........: milters-request@milter.info?subject=remove
More information..: http://www.milter.info/#Support
--------------------------------------------------------

April Lorenzen wrote:

> The admin is still the expert - he determines the mix of factors and
> scoring that fits his own standards, by using a web interface to set how
> his queries will be processed by the Outbound Index. Anthony, as a milter
> writer, has the view that all configuration should be done in the milter,
> and one can alter and update the milter if one wants to make changes.

Thats not entirely accurate. April and I had long discussions about 
interfaces and configuration issues (essentially server implementation 
vs. client implementation and what must pass between).

My objective was to take Petru Paler's original protocol and make a 
independant client/server protocol that did not impose server 
implementation on the client side. This was so as a milter writer I 
could write something that could be used with one or more SIQ supporting 
reputation services. Also as a mail administrator, I wanted the milter 
implementation to be as automated as possible (I've never played with 
the Outbound Index interface to tweak things) chosing to just use the 
defaults.

What can be configured in milter-siq is simply the thresholds at which 
to accept, tag, reject, and/or discard a message. I wanted the client 
side to have at least that much control. The rest of any tweaking 
depends on the service(s) used, such as Outbound Index that provides a 
web front-end on the server for each client to make scoring adjustments.

Also the protocol should provide sufficient enough information (4 scores 
in this draft) such that a client like milter-siq could make some what 
more complex judgements or threshold corrections if necessary.

The SIQ UDP response has room for more things, BUT defining what things 
should go into that response packet that would be of interest to the 
vast majority of reputation services is where I figured would be too 
difficult to get concensus, so I opted for the basics.

Since the UDP packet size is limited as to what can be given back to a 
client, the HTTP quert/response format is provided as a means of getting 
more detailed data concerning the query. HTTP using X- extension headers 
allows for richer response. Though the SIQ protocol only defines a basic 
set, leaving room for server specific extensions.

> The SIQ protocol is flexible enough to support either type of
> implementation - "dumb" milter / query client, or "dumb" query
server.

Yes. "dumb" client works very well, I'm not sure though if the basic 
protocol provides sufficient enough information to allow for "dumb" 
servers that grab a data set and send it back. A server is still 
required to do some additional work to compute scores for the client (at 
least when using UDP packets as discussed above).

-- 
Anthony C Howe                                 +33 6 11 89 73 78
http://www.snert.com/       ICQ:
7116561         AIM: Sir Wumpus

            "Once...we were here."  - Last of The Mohicans


Lists Index Date Thread Search