[Spambayes] Has anyone benchmarked performance?

[Tim Peters]

[Skip Montanaro]
> I'm curious to know if anyone had done any comparative performance
> tests.

Not directly.  Last time I measured spambayes, it scored 80 msgs/second,
wall-clock time, across very large test runs, on an 866MHz P3.  That
included file I/O, tokenization, scoring, and all the hair the test
framework imposed to compute statistics and print out oddball msgs, etc.
This was using an in-memory dict; I'd be surprised if spambayes weren't
I/O-bound when using a disk-based database.

> Here at Northwestern, the folks who administer the mail servers have
> done some experimenting with SpamAssassin (versions and configuration
> unknown to me) and have, at times, been disappointed in its performance.

Disappointment depends on expectation <wink>.  The computations SB does are
very simple compared to those SA does.  Note that SB is trying to solve an
easier problem, though (separating an individual's notions of ham from spam,
not trying to identify spam in an absolute sense).