Shane Harrelson wrote:
However, in an effort to provide some useful performance data for
everyone, I'm making available the results of my testing with the
various suggested insertion algorithms below. The fastest was my
"Hack" version, followed closely by the "Insert/Select/Insert" version
(which DRH agreed would most likely be the fastest "approved"
mechanism), with the "Insert+Ignore/Insert+Select" version proposed by
Dennis Cote, while being simplest, was the slowest.
Windows Embedded
(ms) (%) (ms) (%)
V0 2718 0 30233 0
V1 3014 10.89 32068 6.07
V2 3468 27.59 36275 19.98
Shane,
Thanks for posting your benchmark data. Your testing method is impeccable.
I'm still surprised the two insert method is so much slower. But, with
your 10% collision rate, it is doing a select as opposed to a call to
sqlite_last_insert_rowid for the 90% of rows that don't collide. In then
end, it seems that adds up to substantial time even when the pages are
all cached.
Dennis Cote
-----------------------------------------------------------------------------
To unsubscribe, send email to [EMAIL PROTECTED]
-----------------------------------------------------------------------------
