I was checking some other greylist implementations [1] when I came across an implementation that only stores a 32-bit hash [2] instead of the full tuple. The idea sounded intriguing, and since I'm already calculating a 32-bit CRC (Cyclic Redundancy Check) as part of my protocol, I thought it might be interesting to see just how effective it might be.
Out of 499,846 unique tuples checked, only 28 had duplicate CRCs. I think I could live with that; I could then store 32× the number of tuples in the same amount of memory. Of course, I lose the ability to see the actual tuples, but for a huge ISP (Internet Service Provider) or webhosting company trying to deal with an insane amount of email, that is certainly a viable option (hint, hint—Rob, you listening?)
[1] http://projects.puremagic.com/greylisting/links.html
[2] http://slett.net/spam-filtering-for-mx/exim-greylisting.html