nntp-extensions Anti-spam NNTP extension draft

[Jeff Garzik]

Chris Caputo wrote:
> Hi.  In section 2.2, why is no response sent back?  I am not sure if
> leaving the lock step approach is a good idea.  If latency is an issue, I
> wonder if the following should be done instead or available as an
> alternative.: [...]

That's a good question.

No response is sent back because of how I plan for this to be
implemented on the server side.

I don't intend for there to be a large number of sites sending FILTER
commands (although such is possible).  I envision a few, large anti-spam
servers running their filters full tilt, and then pumping out the
filtering message ids to hundreds (or thousands) of sites at the same
time.

Since we're using TCP here, we know either (a) a message id was sent, or
(b) the connection was lost.  Given that, I didn't see any need to incur
the processing overhead on the FILTER server to keep a queue of
backlogged message ids (in case a negative response is returned).  Once
you start backlogging ids, you add latency (the very thing this is
trying to reduce).  The FILTER server simply sends out message ids to
all connected clients.  If a client (downstream site) can't be
contacted, those FILTER commands to that particular site are simply
lost.

I would agree that a response is necessary if the FILTER feed is a
site's sole means for despamming.  The intention of FILTER is not a full
and complete anti-spam solution, but rather a fast, quick solution that
will beat spam to leaf nodes.
-- 
Jeff Garzik			    Typhoon, Cyclone, Diablo, and INN
Spinne USENET News Service	        Fast and efficient news feeds
http://www.spinne.com/usenet/	           News tuning and consulting