On Wed, 31 Oct 2007, cour...@thefreecat.org wrote:

The point is that all such non-spamming setups (e.g. gmail) would need to be whitelisted for greylisting to work. Otherwise, greylisting will massively delay (possibly to the point of bouncing) mail from multi-homed systems.

Google for it. It's essentially port knocking for SMTP. For example, you only accept the TCP connection on MX3 if the sending server first touched port 25 on MX1 and MX2, in the correct order. MX1 and MX2 always reject, but MX3 selectively accepts or drops/tarpits.

I haven't bothered. nolisting (decoy MX records) + RBLs knocked spam on the head down to 0.1% of where it was. And the delay my mail sees is a few times the ping time from the sender, which will, if it's RFC compliant, retry the next MX until it finds the one that works. Greylisting is a paradigm that is incompatible with this approach.

How much ham bounced (apart from _all_ of it getting delayed by an arbitrary amount of time)? You'll find that nolisting+RBLs approach would have likely yielded at least equivalent results with none of the drawbacks of greylisting, and taken no longer to set up. Nolisting is effectively guaranteed to yield no false positives, and RBLs reject immediately. It is often better to immediately get a bounce as with RBLs than to have the mail sit in limbo with the sender thinking it's been received.

Gordan