I've been reading archives from blogs, postgres mailing lists and here.
After some testing I do seem to get the best results from the
suggestions by Tom Lane. My scheme minus the GREATEST/LEAST [1] is the
same as the suggestions.
1) I'm not certain my sa_bayes scheme is error free (Should I pull from
SVN/snapshot/tarball or is there maybe something not even filed upstream
I should look at?)
2) Based on my benchmarks alone I see positive performance gains, but
still want to ask the list for feedback. [2]
3) It's been discussed to rework the Perl module.. Please pardon me for
not just looking if there's an included TODO, but have most of the
suggestions been a) filed upstream and or b) integrated in the latest
release.
Bottom line question is this : Should I post my scheme based on Tom
Lane's suggestions, make sure that it's solid/error free or is there a
scheme with functions already that I should convert?
Most Kind Regards,
Christopher Bergström
[1] Starting at PostgreSQL 8.1.x has built-in GREATEST/LEAST iirc
[2] After each set of changes I ran the test twice (minus run 6) The
specs on this test box were very low.. AMD Athlon bogomips :
2092.08/256MB ram, and Raptor hd. (, but it's running at udma5 not 6
because I haven't patched the controller module) I've misplaced the link
to the test case, but searching on the postgres mailing lists for
Spamassassin should bring some results. (Or I can just upload.)
1st)
real 13m16.751s
user 0m20.189s
sys 0m3.900s
real 13m18.431s
user 0m20.237s
sys 0m3.912s
2nd)
real 13m11.233s
user 0m19.233s
sys 0m4.004s
real 13m13.561s
user 0m19.269s
sys 0m4.020s
3rd)
real 4m2.857s
user 0m19.513s
sys 0m3.812s
real 4m4.051s
user 0m19.549s
sys 0m3.840s
4th)
real 3m33.875s
user 0m19.773s
sys 0m3.736s
real 3m34.756s
user 0m19.821s
sys 0m3.748s
5th)
real 3m25.737s
user 0m18.709s
sys 0m4.808s
real 3m27.085s
user 0m18.761s
sys 0m4.820s
* Here is where I changed the SQL per Tom Lane's recommendations (I was
too lazy to run twice, sorry)
6th)
real 2m16.640s
user 0m17.589s
sys 0m4.644s
