Matt Kettler wrote:
Yeah, one thing to be aware of is that insertion seems to be the weak point of
the SQL mode. Even with the best settings SQL is slower than even DBM in phase
1, which is all learning.
I'm a little surprised at your result however.. Such a huge difference implies
that SQLite isn't doing nearly as well as Mysql/pgsql. The worst-case phase 1
difference was 436% slower for SQL than SDBM, but your results are on the order
of 18000% slower..
Where SQL seems to shine is in force-expire (phase 3), and it does well in scan
(phase 2 and 5). It also seems to do poorly in the forget (phase 4).
At first you might think force expire and forget should be delete based, and
thus similar, but in reality force-expire is read/delete intensive, but forget
is update intensive, with few records deleted.
Based on that it appears:
SQL does better than DBM at reading SA bayes records
SQL does better than DBM at deleting SA bayes records
SQL does worse than DBM at inserts.
SQL does worse than DBM at updates.
Whereas SDBM does better than DBM in all tests.
how does the (sleepycat) berkely db compare here? is it just like dbm?
