txn performance penalty

[Greg Louis]

Apologies for not noticing earlier, but I just ran the first largeish
batch of messages (600-odd spam) using 0.93.0 through a script that
scores each individually, registers any misclassified messages, then
scores the messages again.  It took at least three times as long
(subjective judgement was 6x) as I have been used to when running this
script daily with 0.92.x.  I may find time to benchmark more accurately
next weekend.

If this holds in general, an easy option to "live dangerously" might be
required.  Many who've "never had a problem with database corruption"
will feel the performance hit is too high a price to pay for the
increment in reliability.  So far I don't see it as problematical in my
low-volume (2,700 messages/day) environment, but the big guys (the ones
who most need txn) will have a bit of evaluating to do... or perhaps we
need to provide [pointers to] db tuning instructions.

-- 
| G r e g  L o u i s         | gpg public key: 0x400B1AA86D9E3E64 |
|  http://www.bgl.nu/~glouis |   (on my website or any keyserver) |
|  http://wecanstopspam.org in signatures helps fight junk email. |